fix: bound indexed full-history reads

[tom-sapletta-com]

Summary

• serialize stale SQLite index rebuilds with a cross-process lock and recheck freshness after acquisition
• serve full ticket JSON directly from indexed canonical payloads without building the historical Pydantic model cache
• serve operational projections from indexed payloads and retain ordering, filters, pagination and counts

Production finding

After PR #24 deployment, hr-control full-history reads raced on a stale index: Planfile reached 3.1 GiB RSS / 88% CPU and logged SQLite database is locked / locking protocol.

Validation

• pytest -q: 370 passed, 6 skipped
• focused index/API/PLOG suite: 59 passed
• Ruff changed modules and git diff --check: pass
• production-data copy: 3,587 tickets / 78.3 MB JSON rendered in 0.74 s, 214 MiB peak RSS, no _ticket_model_cache
• todo2code focused extraction: z-ai/glm-5.2 via OpenRouter/CoreWeave, no degradation, 0 blocking diagnostics (whole-repo trend remains noisy/non-gating)

Planfile ticket: PLF-3776