perf(pre-commit-advisory): skip the Go hook caches when the diff has no Go

[blairham]

Summary

• The advisory job's time is ~99% cache transfer on non-Go PRs: the Go hook caches are ~1.7GB compressed (k5s), costing 40–50s restore per PR run and ~70s save per main push, while a workflows-only diff's hook run takes ~1s (measured on k5s#99: 72–87s job total).
• Adds a Detect Go-relevant changes step (diff endpoints already available; checkout is fetch-depth: 0) matching *.go, go.mod/go.sum, .golangci.*, .pre-commit-config.yaml, and gates the Go cache restore and the main-push warm/save seed steps on it.
• Skipping the seed on a non-Go push is strictly correct, not just cheap: such a push changes no Go analysis input, so the previously saved cache is still exact and the restore-keys prefix serves it regardless of which sha saved it.
• Unavailable diff base (branch-create zeros, force-push) defaults to changed=true — the failure mode is a wasted restore, never a cold Go PR.
• Expected effect: non-Go PR advisory runs drop from ~72–87s to ~25–30s; non-Go main pushes stop re-uploading an identical 1.7GB. Go-touching runs are unchanged.

Test plan

• actionlint green
• Validation PR k5s#100 (closed after validating), run 31885494820: workflows-only diff logged the skip and the advisory job ran in 30s (from 72–87s)
• Run 31885721380 (.go-touching commit): logged the restore path, restored go-hook-caches-Linux-…, golangci hooks ran warm and passed

🤖 Generated with Claude Code