chore: upgrade Go toolchain 1.25.4 → 1.27.0

[kolkov]

Summary

Upgrade Go toolchain from 1.25.4 to 1.27.0 to benefit from three compiler optimizations:

• Known-bits dataflow pass — tracks provable 0/1 bits, folds redundant ops. May eliminate DFA bounds checks on ft[offset]
• Loop-invariant code motion (LICM) — hoists invariants out of loops. 10.17% improvement on regexp/syntax benchmarks (same loop type as our DFA/NFA)
• Switch-to-lookup table — dense switch→table compilation. Benefits our 13-way strategy dispatch

Also: size-specialized malloc for <80 byte objects, Green Tea GC (old GC removed).

Zero breaking changes. All tests pass. Assembly files fully compatible.

Test plan

• go build ./... — clean
• go test ./... — all 11 packages pass
• go vet ./... — clean
• CI: tests on 3 OS (ubuntu, macos, windows)
• CI: benchmark comparison vs main (Go 1.25.4)
• regex-bench: full comparison on AMD EPYC (Go+Rust)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

Benchmark Comparison

Comparing main → PR #165

Summary: geomean 76.34n 76.73n +0.50%

⚠️ Potential regressions detected:

geomean                    76.34n         76.73n        +0.50%
geomean                               ³                +0.00%               ³
geomean                               ³                +0.00%               ³
geomean              32.32n         32.34n        +0.05%
geomean                         ³                +0.00%               ³
geomean                         ³                +0.00%               ³
AnchoredLiteralVsStdlib/stdlib_long-4                   510.0n ± ∞ ¹    554.9n ± ∞ ¹    +8.80% (p=0.008 n=5)
LangArenaLogParserStdlib/errors-4                       9.847m ± ∞ ¹    9.973m ± ∞ ¹    +1.28% (p=0.032 n=5)
IPRegex_Find/stdlib_1KB_sparse-4                        19.38µ ± ∞ ¹    55.01µ ± ∞ ¹  +183.85% (p=0.008 n=5)
IPRegex_Find/stdlib_64KB_sparse-4                       404.7µ ± ∞ ¹   1126.2µ ± ∞ ¹  +178.28% (p=0.008 n=5)

Full results available in workflow artifacts. CI runners have ~10-20% variance.
For accurate benchmarks, run locally: ./scripts/bench.sh --compare