feat(phase-34): add api/routes/case_memory.py with 3 endpoints

GET /case-memory/stats -- case count, pattern breakdown, false positive rates
GET /case-memory/similar -- find past cases with matching finding types
POST /case-memory/false-positive -- record false positives to improve accuracy

CaseStore was fully implemented in ai/case_memory/case_store.py.
It was imported by zero routes and had no feedback mechanism.
The false-positive endpoint closes the learning loop: users flag bad
findings, WeightOptimizer picks them up and reduces the offending
indicator weight on next optimize() run.

api/routes/case_memory.py

@@ -0,0 +1,94 @@
+"""
+BharatGraph - Phase 34: Case Memory API
+GET  /case-memory/stats               -- how many cases stored, pattern counts
+GET  /case-memory/similar             -- find past cases similar to given findings
+POST /case-memory/false-positive      -- record a false positive to improve accuracy
+
+Pure ASCII.
+"""
+import os, sys
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+
+from datetime import datetime
+from typing import List
+from fastapi import APIRouter, HTTPException
+from pydantic import BaseModel
+from loguru import logger
+
+router = APIRouter(prefix="/case-memory", tags=["CaseMemory"])
+
+
+class FalsePositiveRequest(BaseModel):
+    finding_type: str
+    entity_id:    str
+    reason:       str = ""
+
+
+@router.get("/stats")
+def case_memory_stats():
+    """
+    Return how many investigation cases are stored in memory,
+    breakdown by finding type, and false positive rates.
+    """
+    logger.info("[CaseMemory] stats")
+    try:
+        from ai.case_memory.case_store import CaseStore
+        cs = CaseStore()
+        return {
+            "stats":        cs.get_pattern_stats(),
+            "total_cases":  cs.get_case_count(),
+            "analyzed_at":  datetime.now().isoformat(),
+        }
+    except Exception as e:
+        logger.error(f"[CaseMemory] stats error: {type(e).__name__}")
+        return {"status": "error", "detail": str(type(e).__name__)}
+
+
+@router.get("/similar")
+def find_similar_cases(finding_types: str = ""):
+    """
+    Find past investigation cases that share the same finding types.
+    Pass finding_types as a comma-separated list.
+    Example: /case-memory/similar?finding_types=benfords_law_anomaly,ghost_company
+    """
+    logger.info(f"[CaseMemory] similar lookup: {finding_types[:60]}")
+    try:
+        from ai.case_memory.case_store import CaseStore
+        cs = CaseStore()
+        ft_list = [f.strip() for f in finding_types.split(",") if f.strip()]
+        findings = [{"type": ft} for ft in ft_list]
+        similar  = cs.find_similar(findings, top_k=10)
+        return {
+            "query_finding_types": ft_list,
+            "similar_cases":       similar,
+            "analyzed_at":         datetime.now().isoformat(),
+        }
+    except Exception as e:
+        logger.error(f"[CaseMemory] similar error: {type(e).__name__}")
+        return {"status": "error", "detail": str(type(e).__name__)}
+
+
+@router.post("/false-positive")
+def record_false_positive(body: FalsePositiveRequest):
+    """
+    Record that a specific finding type produced a false positive for an entity.
+    This feedback is used by WeightOptimizer to reduce the weight of
+    over-firing indicators.
+    """
+    logger.info(f"[CaseMemory] false positive: {body.finding_type} entity={body.entity_id[:8]}")
+    try:
+        from ai.case_memory.case_store import CaseStore
+        cs = CaseStore()
+        cs.record_false_positive(
+            finding_type=body.finding_type,
+            entity_id=body.entity_id,
+        )
+        return {
+            "recorded":     True,
+            "finding_type": body.finding_type,
+            "entity_id":    body.entity_id[:8] + "...",
+            "recorded_at":  datetime.now().isoformat(),
+        }
+    except Exception as e:
+        logger.error(f"[CaseMemory] false positive record error: {type(e).__name__}")
+        raise HTTPException(status_code=500, detail="Failed to record feedback")