Use Java 17 string and hex fast paths

[He-Pin]

Summary

This PR uses the Java 17 baseline to move a few hot string/encoding paths onto JIT/GC-friendlier implementations while keeping Scala.js and Scala Native behavior covered by platform fallbacks.

• JVM uses String.repeat for repeated string padding / std.repeat and HexFormat for digest-to-hex.
• JS/Native keep Platform.repeatString fallbacks; Native also gets a manual digest-to-hex loop instead of per-byte String.format.
• UTF-8 conversions in touched native helpers now use StandardCharsets.UTF_8 instead of charset names / default charset.
• std.splitLimit advances with String.indexOf, std.resolvePath uses lastIndexOf, and CLI binding parsing avoids regex String.split.
• Adds small reproducible JMH inputs under bench/resources/jdk17_suite as supplemental targeted coverage.
• Adds compact directional tests for repeat/format/split/resolvePath and one JVM/Native Unicode hash vector.

I intentionally did not use Files.readString: its malformed-input behavior is not equivalent to the current new String(Files.readAllBytes(...), UTF_8) decode path. I also left stream/matcher helpers and String.strip/isBlank/lines/formatted out of scope because they either change Jsonnet semantics, add allocation-heavy pipelines, or do not map to the current hot paths.

This is not claimed as a broad existing-suite speedup or as the theoretical limit of all Java 17 APIs. The existing JMH suite is mostly neutral/noisy; the measurable wins are concentrated in the string repeat/split/resolve/hash paths shown below.

JMH

Environment:

• JVM: Zulu OpenJDK 21.0.10 on macOS/aarch64
• Project baseline: current master (3a9a4928, Java 17 compile level)
• Existing-suite flags: -wi 3 -i 5 -w 1s -r 2s -f 1 -jvmArgsAppend -Xss100m -prof gc
• Targeted-suite flags: -wi 3 -i 5 -w 1s -r 2s -f 1 -prof gc

Existing JMH Suite

This run uses the 36 existing bench/resources/{bug_suite,cpp_suite,go_suite,sjsonnet_suite} inputs and excludes the new jdk17_suite files.

metric	result
time geomean	+1.21%
allocation geomean	-0.10%
time count, <= -1% / flat / >= +1%	10 / 10 / 16
allocation count, <= -1% / flat / >= +1%	1 / 35 / 0

Top existing-suite timing movements:

direction	benchmark	base ms/op	opt ms/op	time	alloc
faster	bench/resources/cpp_suite/bench.07.jsonnet	3.124	2.573	-17.6%	-0.01%
faster	bench/resources/cpp_suite/bench.01.jsonnet	0.047	0.045	-3.7%	-0.05%
faster	bench/resources/go_suite/manifestYamlDoc.jsonnet	0.056	0.055	-2.6%	-0.04%
faster	bench/resources/go_suite/manifestTomlEx.jsonnet	0.069	0.068	-2.0%	+0.40%
faster	bench/resources/go_suite/lstripChars.jsonnet	0.115	0.113	-1.7%	-0.00%
slower	bench/resources/cpp_suite/realistic2.jsonnet	39.855	47.023	+18.0%	+0.01%
slower	bench/resources/cpp_suite/bench.02.jsonnet	25.955	29.907	+15.2%	+0.00%
slower	bench/resources/sjsonnet_suite/setInter.jsonnet	0.348	0.379	+9.0%	+0.02%
slower	bench/resources/cpp_suite/large_string_join.jsonnet	0.539	0.583	+8.2%	-0.00%
slower	bench/resources/go_suite/substr.jsonnet	0.055	0.058	+5.4%	-0.00%

Most allocation-positive existing-suite movements:

benchmark	alloc	time
bench/resources/cpp_suite/bench.09.jsonnet	-1.57%	-0.2%
bench/resources/go_suite/escapeStringJson.jsonnet	-0.97%	-1.6%
bench/resources/bug_suite/assertions.jsonnet	-0.79%	+3.5%
bench/resources/go_suite/base64_stress.jsonnet	-0.64%	+1.2%
bench/resources/go_suite/comparison.jsonnet	-0.09%	+0.3%

Interpretation: allocation is effectively flat on the existing suite, with no >= +1% allocation regressions. Timing is mixed and slightly slower on geomean in this single run; the largest slower entries have essentially unchanged allocation, so I am treating the existing-suite result as neutral/noisy rather than a throughput win. I am therefore not claiming an existing-suite throughput improvement from this PR.

Full existing-suite table

benchmark	base ms/op	opt ms/op	time	base B/op	opt B/op	alloc
bench/resources/bug_suite/assertions.jsonnet	0.191	0.198	+3.5%	644,769	639,692	-0.79%
bench/resources/cpp_suite/bench.01.jsonnet	0.047	0.045	-3.7%	80,329	80,289	-0.05%
bench/resources/cpp_suite/bench.02.jsonnet	25.955	29.907	+15.2%	81,680,412	81,681,024	+0.00%
bench/resources/cpp_suite/bench.03.jsonnet	6.722	6.782	+0.9%	5,917,382	5,917,344	-0.00%
bench/resources/cpp_suite/bench.04.jsonnet	0.110	0.109	-1.3%	633,739	633,795	+0.01%
bench/resources/cpp_suite/bench.06.jsonnet	0.217	0.227	+4.5%	878,935	878,969	+0.00%
bench/resources/cpp_suite/bench.07.jsonnet	3.124	2.573	-17.6%	2,970,731	2,970,445	-0.01%
bench/resources/cpp_suite/bench.08.jsonnet	0.039	0.038	-1.4%	81,945	81,953	+0.01%
bench/resources/cpp_suite/bench.09.jsonnet	0.042	0.042	-0.2%	91,217	89,785	-1.57%
bench/resources/cpp_suite/gen_big_object.jsonnet	0.806	0.829	+2.8%	4,015,687	4,015,736	+0.00%
bench/resources/cpp_suite/large_string_join.jsonnet	0.539	0.583	+8.2%	1,530,958	1,530,940	-0.00%
bench/resources/cpp_suite/large_string_template.jsonnet	1.579	1.559	-1.3%	7,776,578	7,776,501	-0.00%
bench/resources/cpp_suite/realistic1.jsonnet	1.415	1.412	-0.3%	5,875,529	5,875,726	+0.00%
bench/resources/cpp_suite/realistic2.jsonnet	39.855	47.023	+18.0%	75,875,290	75,880,789	+0.01%
bench/resources/go_suite/base64.jsonnet	0.145	0.150	+3.7%	2,184,109	2,184,045	-0.00%
bench/resources/go_suite/base64Decode.jsonnet	0.116	0.117	+0.5%	1,547,577	1,547,529	-0.00%
bench/resources/go_suite/base64DecodeBytes.jsonnet	5.017	5.217	+4.0%	232,418	232,580	+0.07%
bench/resources/go_suite/base64_byte_array.jsonnet	0.772	0.782	+1.3%	4,694,249	4,694,403	+0.00%
bench/resources/go_suite/base64_stress.jsonnet	0.175	0.178	+1.2%	842,575	837,150	-0.64%
bench/resources/go_suite/comparison.jsonnet	0.029	0.029	+0.3%	60,617	60,561	-0.09%
bench/resources/go_suite/comparison2.jsonnet	17.200	17.209	+0.1%	37,670,053	37,670,139	+0.00%
bench/resources/go_suite/escapeStringJson.jsonnet	0.032	0.031	-1.6%	77,336	76,584	-0.97%
bench/resources/go_suite/foldl.jsonnet	0.071	0.070	-0.9%	350,674	350,602	-0.02%
bench/resources/go_suite/lstripChars.jsonnet	0.115	0.113	-1.7%	1,533,089	1,533,057	-0.00%
bench/resources/go_suite/manifestJsonEx.jsonnet	0.052	0.051	-0.7%	132,273	132,185	-0.07%
bench/resources/go_suite/manifestTomlEx.jsonnet	0.069	0.068	-2.0%	145,153	145,737	+0.40%
bench/resources/go_suite/manifestYamlDoc.jsonnet	0.056	0.055	-2.6%	135,977	135,921	-0.04%
bench/resources/go_suite/member.jsonnet	0.635	0.630	-0.7%	2,594,716	2,594,889	+0.01%
bench/resources/go_suite/parseInt.jsonnet	0.032	0.032	+0.1%	68,009	68,113	+0.15%
bench/resources/go_suite/reverse.jsonnet	6.765	6.653	-1.7%	11,836,940	11,837,065	+0.00%
bench/resources/go_suite/rstripChars.jsonnet	0.113	0.115	+1.9%	1,532,424	1,532,425	+0.00%
bench/resources/go_suite/stripChars.jsonnet	0.112	0.114	+1.5%	1,523,152	1,523,048	-0.01%
bench/resources/go_suite/substr.jsonnet	0.055	0.058	+5.4%	507,139	507,115	-0.00%
bench/resources/sjsonnet_suite/setDiff.jsonnet	0.386	0.398	+3.1%	1,328,027	1,328,178	+0.01%
bench/resources/sjsonnet_suite/setInter.jsonnet	0.348	0.379	+9.0%	1,266,231	1,266,465	+0.02%
bench/resources/sjsonnet_suite/setUnion.jsonnet	0.582	0.589	+1.3%	1,509,564	1,509,463	-0.01%

Command:

PATHS=$(find bench/resources/bug_suite bench/resources/cpp_suite bench/resources/go_suite bench/resources/sjsonnet_suite -type f -name '*.jsonnet' | sort | paste -sd, -)
./mill --no-server bench.runJmh sjsonnet.bench.RegressionBenchmark.main \
  -p path="$PATHS" \
  -wi 3 -i 5 -w 1s -r 2s -f 1 \
  -jvmArgsAppend -Xss100m \
  -prof gc \
  -rf json -rff /private/tmp/sjsonnet-jdk17-bench/{base,opt}-existing3.json

Supplemental Targeted JMH

These new inputs isolate the paths changed in this PR; they are supplemental and are not a replacement for the existing-suite comparison above.

benchmark	base ms/op	opt ms/op	time	base B/op	opt B/op	alloc
bench/resources/jdk17_suite/hash.jsonnet	1.584	1.611	+1.7%	4,222,936	3,801,302	-9.98%
bench/resources/jdk17_suite/repeat_format.jsonnet	0.270	0.235	-13.1%	741,375	724,943	-2.22%
bench/resources/jdk17_suite/split_resolve.jsonnet	0.203	0.141	-30.3%	846,347	690,568	-18.41%

Targeted geomean:

• time: -14.93%
• allocation: -10.45%

Command:

./mill --no-server bench.runJmh sjsonnet.bench.RegressionBenchmark.main \
  -p path=bench/resources/jdk17_suite/hash.jsonnet,bench/resources/jdk17_suite/repeat_format.jsonnet,bench/resources/jdk17_suite/split_resolve.jsonnet \
  -wi 3 -i 5 -w 1s -r 2s -f 1 \
  -prof gc \
  -rf json -rff /private/tmp/sjsonnet-jdk17-bench/{base,opt}-targeted2.json

Correctness / Tests

./mill --no-server 'sjsonnet.jvm[3.3.7].checkFormat'
./mill --no-server 'sjsonnet.jvm[3.3.7].test'
./mill --no-server 'sjsonnet.js[3.3.7].test'
./mill --no-server 'sjsonnet.native[3.3.7].test'

All passed locally.

[stephenamar-db]

rebase