v0.2.0

🚀 Major Improvements

PERFORMANCE: 46% faster indexing (19.35s → 10.5s on real-world codebases!)

Queue-Based Architecture (NEW!)

Complete rewrite of the indexing pipeline with proper async queues:

• FileContentQueue + ChunkQueue with back pressure
• Parser workers for concurrent file processing
• Embedding workers for parallel GPU batching
• Replaced 222 lines of complex batching logic with clean worker pattern
• Foundation for PostgreSQL queues (Issue #35)

Critical Bug Fixes

• Fixed 5.6x over-embedding bug (chunks re-embedded in loops)
• Fixed 2x over-embedding bug (redundant warmup calls)
• Fixed false truncation warnings
• Fixed chunk_index constraint violations
• Fixed token counting accuracy

Configuration Consolidation

• Read model config from HuggingFace (max_position_embeddings)
• Removed Profile enum complexity
• Consolidated batch size configs (indexer vs search)
• Optimized defaults: pool_size=8, batch_size=1, max_tokens=512
• Single source of truth in .env.sample

Performance Optimizations

• 241 chunks indexed correctly (vs 214 baseline)
• Pool size: 8 workers (optimal parallelism)
• Batch size: 1 (eliminates request batching overhead)
• No chunk_index collisions

Infrastructure

• Added cargo-release workflow for automated version bumping
• Release Drafter for auto-generated changelogs
• Comprehensive timing instrumentation

Full Changelog: v0.1.5...v0.2.0

---

Closes #34