1.3.2: storage bloat remediation, doctor --fix cross-project repair, and patch bump

[LucasArmstrong]

Summary

Release v1.3.2 resolves a critical LanceDB storage bloat issue where stale indices and unpruned version manifests accumulated on disk, growing directories to upwards of 6.4 GB (with 98%+ waste). Across workspace databases, disk usage was reduced from ~7.5 GB to ~19 MB (99.7% reclamation) while keeping 100% of visual states intact.

Key Changes

• Storage Health Engine (src/core/storage-health.ts): Instant non-locking audit of data fragments, stale indices, old version manifests, and transaction logs.
• doctor --fix: Adds automated local database compaction, stale index cleanup, and version manifest pruning to the doctor command.
• doctor-global --fix & --scan <dir>: Added cross-project storage bloat audit and repair across all registered/discovered projects in a single sweep.
• Index Deduplication: Updated src/core/storage.ts to check table.listIndices() before attempting to recreate scalar or vector indexes on startup.
• LanceDB Compaction & Cleanup: Enhanced storage.optimize() with cleanupOlderThan and deleteUnverified: true to purge unreferenced historical files.
• Background Periodic Compaction: Added periodic compaction every 6 eviction sweeps (~30 min) in src/core/eviction.ts and extended shutdown timeout to 5,000ms.
• Size Limit: Lowered default MAX_LANCEDB_SIZE_MB from 1,000 MB to 500 MB.
• Test Suite: Added tests/unit/storage-health.test.ts. All 80 test files (394 tests) pass with full matrix and CI validation.