perf(transform): hash decimal buckets without allocating

[fallintoplace]

What

• Hash decimal bucket values directly from decimal128 signed two-complement bytes.
• Keep DecimalLiteral.MarshalBinary unchanged.
• Add boundary and random compatibility tests.
• Add focused benchmarks.

Why

Decimal bucketing used to build a big.Int and an allocated byte slice for every value.

Benchmark

Apple M1 Pro, arm64. Inputs are pre-boxed:

Path	Before	After	Result
Bucket Apply	124.9 ns, 97 B, 4 allocs	17.3 ns, 0 B, 0 allocs	7.2x faster, zero allocations
Bound Transformer	126.3 ns, 97 B, 4 allocs	13.5 ns, 0 B, 0 allocs	9.4x faster, zero allocations

Tests

• go test ./...
• go vet .