Use bleak-retry-connector and write-only move loop for BLE proxy reliability

[pedropombeiro]

Summary

Replace the raw BleakClient.connect() path with bleak_retry_connector.establish_connection() and rewrite move_to_target to eliminate GATT reads during movement, building on the wakeup ordering fix from #40.

Connection improvements

• add bleak-retry-connector as a dependency
• use establish_connection() with BleakClientWithServiceCache for proper connection caching
• introduce ManagedIdasenDesk subclass to avoid creating an unused BleakClient in __init__ and cleanly adopt the externally-established client
• accept the current BLEDevice in ConnectionManager.connect() so device-swap and reconnect paths stay aligned

Write-only move loop

• send target height writes every 400ms with response=False (fire-and-forget)
• detect arrival via BLE notifications from monitor() instead of polling get_height_and_speed()
• add 15-second safety timeout to prevent infinite move loops

Why

The upstream move_to_target interleaves GATT reads with writes during movement. BLE has a single ATT channel per connection — reads block the channel until the response arrives. Over Bluetooth proxies (HA → WiFi → ESP32 → BLE → desk), read latency is 50–200ms normally but can spike to seconds on timeout/retry, creating unpredictable gaps between reference input writes. The desk interprets these gaps as "stop commanding movement" and decelerates or stops.

By removing reads from the loop, the write cadence becomes deterministic and the ATT channel stays free for outgoing commands. Height data is already pushed to us via the existing monitor() BLE notification subscription.

Ref: #38 (comment: #38 (comment))

Validation

• verified locally against Home Assistant + ESPHome Bluetooth proxy
• confirmed the BleakClient.connect() called without bleak-retry-connector warning is gone
• confirmed the desk reconnects with ESPHome proxy logs showing Connecting v3 with cache
• confirmed smooth desk movement without interruptions over BLE proxy
• 51 tests pass

Supersedes #39. Related to home-assistant/core#155912.

[abmantis]

Hey! This is doing too much for a single PR.
Can you please split each feature/fix/change into separate PRs? Also, have you tried suggesting the improvements to IdasenDesk upstream?
It would be cleaner than extending the class here like that, and could be an improvement to the upstream library.

[pedropombeiro]

Hey! This is doing too much for a single PR. Can you please split each feature/fix/change into separate PRs? Also, have you tried suggesting the improvements to IdasenDesk upstream? It would be cleaner than extending the class here like that, and could be an improvement to the upstream library.

Thanks for the review!

I've split the move-loop change off and proposed it upstream in newAM/idasen#489 — that's a small, self-contained fix that makes move_to_target tolerant of transient speed == 0 readings and reduces the write interval. Once that lands and is released, the desk-stuttering symptom is fixed at the source, and this PR can drop the move_to_target work entirely.

That leaves this PR scoped to the BLE connection path:

• add bleak-retry-connector as a dependency
• use establish_connection() with BleakClientWithServiceCache
• introduce ManagedIdasenDesk only to skip the unused BleakClient in __init__ and adopt an externally-established client via set_client()
• accept the current BLEDevice in ConnectionManager.connect() so device-swap and reconnect paths stay aligned

I'll trim this PR down to just the connection-path changes once we agree.

[abmantis]

introduce ManagedIdasenDesk only to skip the unused BleakClient in init and adopt an externally-established client via set_client()

I just don't like this part very much 😄 . Can we open another PR to upstream, adding an argument client : BleakClient | None = None to IdasenDesk.__init__()? If it is None, it keeps the existing behavior (create a BleakClient). Otherwise the code gets very fragile to changes in upstream.

[pedropombeiro]

Good call — that's a cleaner design. I'll open a follow-up upstream PR on newAM/idasen to add client: BleakClient | None = None to IdasenDesk.__init__(). When it's None, the existing behavior is preserved; otherwise the injected client is used as-is.

Once that lands and is released, I'll:

• drop ManagedIdasenDesk from this PR entirely (use IdasenDesk(..., client=injected_client) directly in ConnectionManager)
• drop the move-loop changes (already fixed upstream in newAM/idasen#489, released as idasen 0.13.0)
• bump the idasen floor to whichever upstream version contains the client argument

Will update here once the upstream PR is filed.