test(search): replay raw intent matches in golden eval

[tae2089]

Summary

• store raw matched intent documents in the golden fixture and replay them through the current intentrank.Rank
• normalize node/document/query data, sort references, and fail fast on duplicate, dangling, unreachable, or cross-pool identity conflicts
• recapture name and intent fixtures from one graph snapshot and structurally rebaseline diagnostics without changing judgments or headline metrics

Why

The previous golden replay froze output after intent ranking, so scorer changes could not affect make search-eval. This restores a red-capable evaluation path for #95 while keeping production ranking behavior unchanged.

Two general ranking rules were evaluated for issue 95, but both regressed other corpus metrics and were reverted. This PR therefore prepares valid measurement while leaving the ranking-debt issue open.

Validation

• CGO_ENABLED=1 go test -tags "fts5" ./... -count=1
• CGO_ENABLED=1 go vet -tags fts5 ./...
• make search-eval across ccg, cobra, context-diary, and gorm
• cross-pool and same-query identity collision checks: zero conflicts
• ccg build .
• ccg docs --out docs
• ccg lint

PostgreSQL parity was not rerun because no disposable PostgreSQL instance was available; the protected user-owned instance was not used.