feat: Roadmap K — RAG corpus management with score-based eviction (v0.13.0)

- Config: RAG_MAX_VECTORS, high/low water marks, grace period, weights
- rag.py: get_collection_count, evict_batch, maybe_evict (asyncio.Lock),
  get_rag_operational_stats, EVICTION_LOG, retrieval_count tracking
- routers/rag_admin.py: GET /api/rag/stats, POST /api/rag/flush (admin)
- Wire maybe_evict() into upload.py and ingest.py after Qdrant upsert
- 16 tests: collection stats, eviction scoring, pinned/grace/batch guards,
  endpoint auth, race lock, flush, operational stats shape
- Bump to v0.13.0

config.py

@@ -9,7 +9,7 @@ import logging
 
 log = logging.getLogger("jarvischat")
 
-VERSION = "v0.11.0"
+VERSION = "v0.13.0"
 OLLAMA_BASE = os.environ.get("OLLAMA_BASE", "http://localhost:11434")
 LLAMA_SERVER_BASE = os.environ.get("LLAMA_SERVER_BASE", "http://192.168.50.108:8081")
 SEARXNG_BASE = "http://localhost:8888"
@@ -53,6 +53,16 @@ SUPPORTED_UPLOAD_TYPES = {"text/plain", "text/markdown", "application/pdf", "app
 UPLOAD_CONTEXT_EXPIRY_HOURS = 1
 BODY_LIMIT_UPLOAD_BYTES = MAX_UPLOAD_BYTES
 
+# --- RAG eviction ---
+RAG_MAX_VECTORS = 50000
+RAG_EVICTION_HIGH_WATER = 0.80
+RAG_EVICTION_LOW_WATER = 0.20
+RAG_EVICTION_BATCH = 1000
+RAG_PINNED_SOURCES = ["upload", "profile"]
+RAG_GRACE_HOURS = 1
+RAG_ACCESS_WEIGHT = 1.0
+RAG_AGE_WEIGHT = 0.1
+
 MAX_CHAT_MESSAGE_CHARS = 8000
 MAX_SEARCH_QUERY_CHARS = 500
 MAX_PROFILE_CHARS = 32000