fix: replace BleakClient.connect() with establish_connection() and add exponential backoff

[bpinson999]

Closes #5

What this fixes
When the vaporizer is unplugged or out of range, the integration spammed the HA log with 126,000+ warning-level entries in a ~2 hour window. See #5 for the full problem description.

Changes
bluetooth_coordinator.py

• Replace BleakClient.connect() with bleak_retry_connector.establish_connection(), resolving the habluetooth.wrappers warning about unreliable connection establishment
• Use async_ble_device_from_address() to check HA's Bluetooth registry before attempting a connection — if the device isn't advertising, we return immediately with a DEBUG log instead of a WARNING
• Catch BleakNotFoundError separately so a powered-off device logs at DEBUG level rather than WARNING
• Add exponential backoff in _run() (5s base, doubles each failure, 300s cap) to bound retry frequency when the device is unavailable long-term
• Use asyncio.wait_for(stop_event.wait()) for the backoff sleep so stop() wakes the loop immediately rather than waiting out the full delay
• Add _on_disconnect callback for clean state reset on unexpected disconnection
• Accept hass as a constructor argument (required for async_ble_device_from_address)

__init__.py

• Pass hass into VolcanoBTManager constructor

Tested
Verified with the device unplugged. Backoff progression confirmed in logs (5s → 10s → 20s → ...). No habluetooth.wrappers warnings observed. Zero log spam.

AI Disclosure
This fix was developed with assistance from Claude. All changes were reviewed, tested, and verified by me on a live Home Assistant instance.