fix: apply PR cache in daemon hot paths to prevent rate-limit exhaustion

[proboscis]

Summary

• Make LookupInfoWithClient and LookupInfoByURL cache-aware in internal/pr/cache.go by loading/saving cache entries, honoring cacheHitTTL/cacheMissTTL, and applying cacheMinFetchInterval.
• Add cache-first wrappers (LookupInfoCached, LookupInfoByURLCached) and switch daemon hot paths to them in internal/daemon/monitor.go, internal/daemon/types.go, and internal/daemon/proto_handler.go.
• Remove orch-450 semgrep exclusions for daemon files in .semgrep/architecture.yaml section 11 so architecture lint now enforces this path directly.

Verification

• go test ./internal/pr/... ✅
• go test ./internal/daemon/... ✅
• go test -count=1 ./internal/pr/... ./internal/daemon/... ✅
• make lint ✅ (0 findings)

Acceptance Criteria Evidence

1. LookupInfo / LookupInfoByURL use cache + TTL/min interval

   • Implemented in internal/pr/cache.go via cache read/write helpers and min-fetch gating.
   • TDD tests now pass:
     • TestLookupInfoWithClient_MustUseCache
     • TestLookupInfoWithClient_CacheHitSkipsAPI
     • TestLookupInfoByURL_MustUseCache
     • TestLookupInfoByURL_CacheHitSkipsAPI

2. checkPRMergedWithURL throttled (not every 5s API call)

   • internal/daemon/monitor.go now calls cached wrappers only.
   • No direct pr.LookupInfo* calls remain in daemon package; semgrep architecture rule enforces this.

3. lookupPRInfo in types.go uses cached data

   • internal/daemon/types.go now routes through LookupInfoCached / LookupInfoByURLCached.

4. Consider global rate limiter

   • Not added in this change; existing daemon hot-loop bypasses are removed and now governed by cache throttling as a safety baseline.

5. 20 active runs under 3000 calls/hr

   • Worst-case miss path bounded by cacheMissTTL=30s: 20 runs * (3600/30) = 2400 calls/hr (below 3000/hr target).
   • Cache hits are substantially lower.

6. No regression to existing cache TTL constants / PopulateRunInfo

   • Existing constants unchanged.
   • PopulateRunInfo path still passes regression tests in internal/pr/cache_test.go.

7. All new TDD tests pass

   • Confirmed by go test ./internal/pr/....

8. Semgrep exclusions removed

   • Removed for monitor.go, types.go, proto_handler.go in .semgrep/architecture.yaml section 11.
   • Confirmed by make lint with 0 findings.

Refs: orch-450