Fix credential_db schema conflict and add resilience

credential_sniffer and credential_db both wrote to the same credentials.db
with different schemas; credential_db's executescript failed when creating
the crack_status index on an existing table that lacked that column.

Fix: split credential_db's schema init into table DDL + migration + indexes
so missing columns are added before index creation runs. Migration also
handles any future schema drift.

state.py: add 30s timeout + busy_timeout PRAGMA to write connection so
concurrent subprocess writes stop failing with 'database is locked'.

capture_bus.py: reconnect on ENETDOWN instead of aborting the capture
loop — mac_manager temporarily brings the interface down during MAC
rotation which was silently killing all packet capture.

core/capture_bus.py

@@ -170,9 +170,21 @@ class CaptureBus:
             except socket.timeout:
                 continue
             except OSError as e:
-                if self._running:
-                    logger.error("Socket read error: %s", e)
-                break
+                if not self._running:
+                    break
+                logger.warning("Socket read error: %s — reconnecting in 3s", e)
+                try:
+                    self._sock.close()
+                except Exception:
+                    pass
+                time.sleep(3)
+                try:
+                    self._sock = self._open_socket()
+                    logger.info("CaptureBus reconnected on %s", self.interface)
+                except OSError as e2:
+                    logger.error("CaptureBus reconnect failed: %s", e2)
+                    time.sleep(10)
+                continue
 
             ts = time.time()
             self._total_packets += 1