feat(whatsapp): QWEN primary AI backend, Hermes fallback, conversation history, vehicle persistence, demo prompts

- Add QWEN (qwen3.6) as primary AI backend with short system prompt
- Hermes remains as fallback with 45s timeout
- Increase QWEN timeout to 35s, max_tokens to 4000
- Add conversation history loading from whatsapp_messages (last 4 msgs)
- Persist detected vehicle in whatsapp_sessions table
- Add 'limpiar chat' / 'nuevo chat' / 'reset' commands to clear history
- Fix CSS conflict: rename whatsapp chat-panel classes to wa-chat-panel
- Fix JS ID conflicts with chat.js widget (waChatPanel, waChatMessages, etc.)
- Improve no-stock response: conversational with alternatives
- Split search_query by | for multi-part lookups
- Add DEMO_PROMPTS.md and DEMO_PROMPTS_V2.md

pos/tenant_db.py

@@ -1,29 +1,24 @@
-# /home/Autopartes/pos/tenant_db.py
-"""Tenant DB connection manager with pooling.
+"""Tenant DB connection manager.
 
-Uses psycopg2.pool.ThreadedConnectionPool for both master and tenant DBs.
-Connections are returned to the pool on .close() via a thin wrapper —
-this keeps the rest of the codebase unchanged.
+Master DB: creates a fresh connection each time (very light load thanks to
+ tenant_id → db_name cache, so we only hit master ~once per 5 min).
+Tenant DBs: use psycopg2.pool.ThreadedConnectionPool with maxconn=20.
 """
 
+import time
+import threading
 import psycopg2
 from psycopg2 import pool
 from config import MASTER_DB_URL, TENANT_DB_URL_TEMPLATE
 
 
-# ─── Pools ─────────────────────────────────────
-_master_pool = None
+# ─── Tenant Pools ──────────────────────────────
 _tenant_pools = {}
 
-
-def _get_master_pool():
-    """Lazy-initialize master DB connection pool."""
-    global _master_pool
-    if _master_pool is None:
-        _master_pool = pool.ThreadedConnectionPool(
-            minconn=2, maxconn=20, dsn=MASTER_DB_URL
-        )
-    return _master_pool
+# ─── Tenant cache ──────────────────────────────
+_tenant_cache = {}
+_tenant_cache_ttl = 300
+_tenant_cache_lock = threading.Lock()
 
 
 def _get_tenant_pool(db_name):
@@ -37,6 +32,34 @@ def _get_tenant_pool(db_name):
     return _tenant_pools[db_name]
 
 
+def _resolve_tenant_db(tenant_id):
+    """Return db_name for tenant_id, using cache first."""
+    now = time.time()
+    with _tenant_cache_lock:
+        entry = _tenant_cache.get(tenant_id)
+        if entry and entry['expires'] > now:
+            return entry['db_name']
+
+    # Cache miss or expired — query master DB with a fresh connection
+    conn = psycopg2.connect(MASTER_DB_URL)
+    try:
+        cur = conn.cursor()
+        cur.execute(
+            "SELECT db_name FROM tenants WHERE id = %s AND is_active = true",
+            (tenant_id,)
+        )
+        row = cur.fetchone()
+        cur.close()
+        db_name = row[0] if row else None
+    finally:
+        conn.close()
+
+    if db_name:
+        with _tenant_cache_lock:
+            _tenant_cache[tenant_id] = {'db_name': db_name, 'expires': now + _tenant_cache_ttl}
+    return db_name
+
+
 class _PooledConnection:
     """Thin wrapper that delegates all attribute access to the real
     psycopg2 connection, but returns it to the pool on .close().
@@ -52,19 +75,17 @@ class _PooledConnection:
 
     def close(self):
         try:
-            # Rollback any aborted transaction before returning to pool.
-            # Without this, failed transactions leave connections in
-            # 'idle in transaction (aborted)' state, eventually exhausting
-            # the pool.
             if self._conn:
                 try:
                     self._conn.rollback()
                 except Exception:
                     pass
-            self._pool.putconn(self._conn)
+                self._pool.putconn(self._conn)
         except Exception:
-            # If pool is already closed, fall back to real close
-            self._conn.close()
+            try:
+                self._conn.close()
+            except Exception:
+                pass
 
     def __enter__(self):
         return self
@@ -76,27 +97,19 @@ class _PooledConnection:
 # ─── Public API ────────────────────────────────
 
 def get_master_conn():
-    """Get a pooled connection to the master DB."""
-    p = _get_master_pool()
-    return _PooledConnection(p.getconn(), p)
+    """Get a direct connection to the master DB (no pool).
+    
+    Caller MUST close() the connection when done.
+    """
+    return psycopg2.connect(MASTER_DB_URL)
 
 
 def get_tenant_conn(tenant_id):
     """Get a pooled connection to a tenant's DB."""
-    master = get_master_conn()
-    cur = master.cursor()
-    cur.execute(
-        "SELECT db_name FROM tenants WHERE id = %s AND is_active = true",
-        (tenant_id,)
-    )
-    row = cur.fetchone()
-    cur.close()
-    master.close()
-
-    if not row:
+    db_name = _resolve_tenant_db(tenant_id)
+    if not db_name:
         raise ValueError(f"Tenant {tenant_id} not found or inactive")
 
-    db_name = row[0]
     p = _get_tenant_pool(db_name)
     return _PooledConnection(p.getconn(), p)