v0.2.0 — SQLite Cache + Semantic Search

Adds local caching, vector search, and bulk data loading.

New commands:

• gchat load 720h — bulk load + cache + embed last 30 days of data with progress bars
• gchat search 'query' — semantic vector search using nomic-embed-text v1.5 embeddings
• gchat search --keyword 'text' — FTS5 keyword search fallback
• gchat cache stats — show cache row counts, DB size, model status
• gchat cache clear — wipe cached data (--models to also delete embedding model)

Under the hood:

• SQLite cache for conversations, messages, users, memberships
• sqlite-vec for vector similarity search (768-dim embeddings)
• Hugot pure Go backend for offline embedding inference
• nomic-embed-text v1.5 model auto-downloads from HuggingFace on first use (~138MB)
• FTS5 for keyword search
• --since accepts durations (168h), dates (2024-01-15), or RFC3339 datetimes
• All existing commands now automatically cache results on every API call

Binary: gchat-darwin-arm64 (macOS Apple Silicon, 21MB)