fix: ChromaDB provisioner reliability + single-instance

[itstauq]

Summary

Production debugging revealed ChromaDB's internal SQLite (chroma.sqlite3) was corrupted (0 bytes) by process-wick SIGKILL'ing the process group mid-write, causing cascading failures on subsequent boots.

• ChromaDB startup reliability: Kill orphaned process on health check timeout, detect early subprocess exit (fail fast), post-health liveness guard against wrong process on same port
• Graceful shutdown: Extended from 2s fixed sleep to 10s polling loop to reduce SIGKILL corruption risk
• 0-byte DB recovery: Delete corrupted chroma.sqlite3 on startup so ChromaDB recreates it fresh
• Ingestion error annotation: Prefix errors with [dataset_type] for easier debugging
• Single-instance: Add tauri-plugin-single-instance to prevent multiple app instances; second launch focuses existing window
• Window management: Hide from taskbar when minimized to tray, restore on show; log window operation errors instead of silently ignoring

Test plan

• Block port 8100, start app — verify ChromaDB process is killed after timeout, not left orphaned
• Create 0-byte ~/.syft-space/chromadb/chroma.sqlite3, start app — verify it's deleted and ChromaDB starts fresh
• Trigger ingestion error — verify error message shows [local_file] Ingestion error: ... prefix
• Launch app, launch again — verify second launch focuses existing window (including if minimized)
• Close window, verify it disappears from taskbar; reopen via tray/dock, verify it reappears