support Go1.27 regabi

[visualfc]

No description provided.

[fennoai]

Review: Go 1.27 support

Reviewed the hand-written changes; the ~240k lines of generated *_regabi_go127_*.s assembly were assessed at the mechanism level, not line-by-line.

Verified correct

• float32reg_arch.go build tag (ppc64 || ppc64le || riscv64 || s390x) is the exact complement of float32reg_generic.go and pairs with the existing .s implementations — no arch double-declared or missing.
• !go1.27 constraints added consistently across all icall size dirs (icall{512..20480}, internal/icall512, cmd/icall_gen/_data); each build dir has the full set of 5 new go127_* files with matching constraints.
• The runtime·spillArgs/runtime·unspillArgs → local ·spillArgs/·unspillArgs rewrite pairs correctly with the appended helper bodies; MakeFunc call shape is unchanged (no per-call overhead).
• .github/workflows/go.yml: 1.27.x present in all matrices, no leftover guards; actions pinned, no untrusted-input interpolation.
• README.md version-range update is accurate; no dangling #build-linkname-mode anchor.

No security or performance concerns. One robustness note inline.

[fennoai]

Fragile build-tag stripping — silent corruption on unexpected input.

asm = asm[strings.Index(asm, "\n\n")+2:] is guarded only by HasPrefix(asm, "//go:build"). It's correct for all 5 current templates, but fails silently otherwise:

• If \n\n is absent, strings.Index returns -1 and +2 yields index 1, slicing off just the first byte instead of the header.
• If a template's //go:build header isn't followed by a blank line, the first blank line inside the body is matched, truncating real assembly.

Both cases produce a subtly broken .s file while generation still "succeeds." Consider asserting the index is >= 0 and returning an error otherwise (or stripping only the contiguous leading comment/build-tag lines), plus a one-line comment documenting the required template format.