Fix FBNS cleanup after socket failures

[subzeroid]

Summary

• mirror the aiograpi FBNS socket-failure cleanup fix in the sync client
• mark FbnsClient.connected false when FBNS transport send/read fails
• make FBNS disconnect best-effort so a broken socket still closes the transport
• always clear Client.fbns after fbns_disconnect(), even if low-level cleanup raises

This fixes the broken-state/reconnect part reported in subzeroid/aiograpi#396. It does not claim Instagram MQTT sockets remain connected indefinitely; long-lived workers should still reconnect after remote/network disconnects.

Tests

• PYTHONDONTWRITEBYTECODE=1 .venv/bin/pytest -q tests/regression/test_fbns.py::test_fbns_read_once_marks_client_disconnected_when_socket_closes tests/regression/test_fbns.py::test_fbns_disconnect_clears_client_state_after_broken_socket
• PYTHONDONTWRITEBYTECODE=1 .venv/bin/ruff check instagrapi/realtime/fbns.py instagrapi/mixins/realtime.py tests/regression/test_fbns.py
• PYTHONDONTWRITEBYTECODE=1 .venv/bin/pytest -q tests/regression/test_fbns.py tests/regression/test_realtime.py
• PYTHONDONTWRITEBYTECODE=1 .venv/bin/pytest -q tests/regression
• git diff --check

Live smoke on sk.test against this branch:

• session-backed account_info succeeded
• direct FBNS MQTT connect/register succeeded
• fbns_ping returned True
• fbns_disconnect cleared client state

Note: the pool account HTTP proxies returned 407 for MQTT CONNECT, so the successful live smoke used direct MQTT transport on sk.test.