refactor(runtime-catalog): index catalog lookups YEF-947

[Yevanchen]

Summary

• Precompute lookup indexes for runtime catalog entries, public runtime entries, vendors, and preset model identities.
• Replace repeated catalog scans in exported getter APIs with map lookups while preserving first-match behavior and fresh array results from list APIs.
• Add runtime-catalog tests for the indexed helper behavior.

Why

• Closes YEF-947.
• The complexity scan found repeated runtime catalog lookup patterns. This changes key catalog getters from repeated O(n) scans to O(1) map lookups, and changes public vendor derivation from nested O(vrrv) scans to O(r*rv + v) module initialization.

Verification

• Commands:
  • python3 /Users/evanmore/multica_workspaces_desktop-api.multica.ai/baf2a510-889b-4b05-b7c8-c079f5c28938/12b40141/codex-home/skills/complexity-optimizer/scripts/analyze_complexity.py mosoo --format json
  • just fmt-check-path pkgs/runtime-catalog
  • just test-package @mosoo/runtime-catalog
  • just tc-package @mosoo/runtime-catalog
  • bun run --filter @mosoo/runtime-catalog lint
  • just commit-check
• Manual steps: N/A
• Not run:
  • just check was attempted. It passed formatting, docs, lint, typecheck, and most tests, then failed in apps/driver/tests/acp-file-system.test.ts on a local macOS temp-path mismatch (/tmp/... expected, /private/tmp/... received) plus an extra sourceEventId; this is outside the changed package.

Impact

• User/API/contract changes: None.
• Generated files / GraphQL / DB / lockfile: None.
• Env or config changes: None.
• Risk and rollback: Low; revert the single commit to restore the previous linear-scan implementation.

Review

• Closest review areas: pkgs/runtime-catalog/src/runtime-catalog.ts, pkgs/runtime-catalog/tests/runtime-model-identity-admission.test.ts
• Known trade-offs: The indexes are built once at module load; memory overhead is small and bounded by the static generated catalog size.

[Yevanchen]

Review verdict: no demonstrated performance need yet. The current catalog contains only 4 runtimes, 10 vendors, and 32 preset models, while this adds multiple indexes and about 80 source lines. A complexity scan proves asymptotics, not user or runtime impact. Please provide a trace or benchmark from a real hot caller; otherwise the smaller linear lookups should stay.