Initial commit: Vera-AI v2 with async Qdrant, singleton pattern, monthly curation, and configurable UID/GID/TZ

Features:
- AsyncQdrantClient for non-blocking Qdrant operations
- Singleton pattern for QdrantService
- Monthly full curation (day 1 at 03:00)
- Configurable UID/GID for Docker
- Timezone support via TZ env var
- Configurable log directory (VERA_LOG_DIR)
- Volume mounts for config/, prompts/, logs/
- Standard Docker format with .env file

Fixes:
- Removed unused system_token_budget
- Added semantic_score_threshold config
- Fixed streaming response handling
- Python-based healthcheck (no curl dependency)

config/config.toml

@@ -0,0 +1,21 @@
+[general]
+ollama_host = "http://10.0.0.10:11434"
+qdrant_host = "http://10.0.0.22:6333"
+qdrant_collection = "memories"
+embedding_model = "snowflake-arctic-embed2"
+debug = false
+
+[layers]
+# Note: system_token_budget removed - system prompt is never truncated
+semantic_token_budget = 25000
+context_token_budget = 22000
+semantic_search_turns = 2
+semantic_score_threshold = 0.6
+
+[curator]
+# Daily curation: processes recent 24h of raw memories
+run_time = "02:00"
+# Monthly full curation: processes ALL raw memories
+full_run_time = "03:00"
+full_run_day = 1  # Day of month (1st)
+curator_model = "gpt-oss:120b"