[Pratt Parser] Cache source content size in Lexer to avoid repeated variant dispatch in character loops.

[copybara-service]

[Pratt Parser] Cache source content size in Lexer to avoid repeated variant dispatch in character loops.

SourceContentView::size() is defined out-of-line and dispatches via absl::visit over a 4-element absl::variant. Because the character scanning loops in Lexer also invoke out-of-line methods on content_, the compiler cannot hoist content_.size() out of loop conditions, causing it to be re-evaluated on every character scanned.

Caching content_size_ in Lexer improves Pratt parser CPU time across pratt_parser_benchmark by ~5.2% geomean (~6.7% geomean on valid expressions).

Measured with pratt_parser_benchmark (built -c opt --dynamic_mode=off). "Pratt before" is this CL's parent (p4head), so the last column is this CL's own contribution. The ANTLR column is the series baseline and is unaffected by this change.

CPU time, median across 36 interleaved repetitions:

Case	ANTLR	Pratt before	Pratt after	Pratt vs ANTLR	Delta this CL
ParseCommon	210,982 ns	11,606 ns	11,003 ns	19.2x faster	-5.2%
ParseArithmeticChain/10	55,289 ns	2,781 ns	2,608 ns	21.2x faster	-6.2%
ParseArithmeticChain/50	291,952 ns	13,794 ns	13,265 ns	22.0x faster	-3.8%
ParseArithmeticChain/100	559,317 ns	27,506 ns	25,724 ns	21.7x faster	-6.5%
ParseLogicalChain/10	43,318 ns	3,306 ns	2,986 ns	14.5x faster	-9.7%
ParseLogicalChain/50	207,612 ns	15,118 ns	13,940 ns	14.9x faster	-7.8%
ParseLogicalChain/100	442,381 ns	30,089 ns	28,675 ns	15.4x faster	-4.7%
ParseMemberChain/10	34,693 ns	1,819 ns	1,705 ns	20.4x faster	-6.3%
ParseMemberChain/50	161,104 ns	8,633 ns	7,889 ns	20.4x faster	-8.6%
ParseMemberChain/100	335,750 ns	16,917 ns	15,724 ns	21.4x faster	-7.1%
ParseNestedParentheses/10	28,084 ns	803 ns	752 ns	37.3x faster	-6.3%
ParseNestedParentheses/50	123,579 ns	2,841 ns	2,620 ns	47.2x faster	-7.8%
ParseCommonSyntaxErrors	429,515 ns	18,317 ns	17,561 ns	24.5x faster	-4.1%
ParseArithmeticChainSyntaxError/10	72,212 ns	3,256 ns	3,078 ns	23.5x faster	-5.5%
ParseArithmeticChainSyntaxError/50	295,287 ns	14,321 ns	13,419 ns	22.0x faster	-6.3%
ParseArithmeticChainSyntaxError/100	628,385 ns	28,006 ns	26,414 ns	23.8x faster	-5.7%
ParseLogicalChainSyntaxError/10	58,855 ns	3,776 ns	3,553 ns	16.6x faster	-5.9%
ParseLogicalChainSyntaxError/50	235,134 ns	16,104 ns	15,147 ns	15.5x faster	-5.9%
ParseLogicalChainSyntaxError/100	458,219 ns	31,111 ns	29,472 ns	15.5x faster	-5.3%
ParseMemberChainSyntaxError/10	50,335 ns	2,144 ns	2,051 ns	24.5x faster	-4.3%
ParseMemberChainSyntaxError/50	182,649 ns	8,962 ns	8,516 ns	21.4x faster	-5.0%
ParseMemberChainSyntaxError/100	346,654 ns	17,161 ns	16,230 ns	21.4x faster	-5.4%
ParseNestedParenthesesSyntaxError/10	241,350 ns	6,098 ns	5,982 ns	40.3x faster	-1.9%
ParseNestedParenthesesSyntaxError/50	1,561,548 ns	28,725 ns	27,390 ns	57.0x faster	-4.6%
ParseRepeatedSyntaxErrors/10	170,609 ns	6,218 ns	6,044 ns	28.2x faster	-2.8%
ParseRepeatedSyntaxErrors/50	172,554 ns	8,602 ns	8,620 ns	20.0x faster	+0.2%
ParseRepeatedSyntaxErrors/100	180,960 ns	9,324 ns	9,532 ns	19.0x faster	+2.2%
Geomean				22.5x faster	-5.2%