feat: implement stage 1 candidate retrieval and scoring logic using Qdrant and multi-factor heuristics

backend/src/matching/stage1.py

@@ -1,6 +1,6 @@
 from typing import Any
 from qdrant_client import QdrantClient
-from qdrant_client.models import Filter, FieldCondition, MatchValue, Range
+from qdrant_client.models import Filter, FieldCondition, MatchValue
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy import select
 
@@ -20,8 +20,9 @@ def _build_qdrant_filter(jd: dict, session_id: str | None) -> Filter | None:
     conditions = []
     if session_id:
         conditions.append(FieldCondition(key="session_id", match=MatchValue(value=session_id)))
-    if jd.get("min_yoe") is not None:
-        conditions.append(FieldCondition(key="years_of_experience", range=Range(gte=max(0, jd["min_yoe"] - 2))))
+    # NOTE: YOE is intentionally NOT pre-filtered here.
+    # Under-qualified candidates are penalised via the `yoe` component score (weight 0.15)
+    # instead of being silently excluded from results entirely.
     if not conditions:
         return None
     return Filter(must=conditions)