fix(bt): don't idle-disconnect while audio is playing to the controller

[gunCannoli]

Problem

Idle Disconnect drops the controller after its timeout based only on the
absence of meaningful HID input. Audio playback is not considered.

Using the controller as a headset -- a headset plugged into the 3.5 mm jack,
watching a movie or listening to music -- means no button input for long
stretches. The controller idle-disconnects mid-playback and the audio cuts
out. The only workarounds are disabling Idle Disconnect or setting an
impractically long timeout.

Change

One condition in src/bt.cpp, in the inactivity check in
l2cap_packet_handler_cold():

    // Inactivity detection.
    if (mute[1]) { // Microphone mute is enabled.
        return;
    }
    if (audio_output_route_protected()) {
        inactive_time = now_us;
        return;
    }
    if (!meaningful_input_activity && now_us - inactive_time > idle_disconnect_timeout_us()) {
        ...
    }

audio_output_route_protected() (audio_recent() || usb_speaker_streaming_active())
already exists in the same file and is already used for the RSSI idle gate
and output-route protection. This reuses it rather than adding a new signal.

The idle clock is reset (rather than simply returning) so the full timeout
restarts when playback stops -- matching how a button press resets it.
Returning without the reset would leave a long playback session about to trip
the timeout the moment audio ends.

Placement is after the existing mute[1] mic-mute carve-out and before the
timeout comparison. It mirrors that carve-out in location and shape.

Scope

• No new state, no timer, no allocation.
• No companion command, no protocol change, no COMMAND_ID, no
  PROTOCOL_MINOR bump.
• No UI change. The existing Idle Disconnect enable toggle and timeout
  selector behave exactly as before; they simply stop firing during
  playback.
• Runs only on HID interrupt packets, the same cadence as the existing
  check.

Testing

tests/firmware/usb_descriptor_migration_test.cpp -- the RSSI-idle source
assertion is extended to require the new guard, its exact body, and its
ordering between the mic-mute carve-out and the timeout comparison. All
firmware host-side suites pass.

Hardware (Waveshare RP2350B-Plus-W, Idle Disconnect set to 1 minute):

1. Headset in the jack, continuous playback, no controller input -- stays
   connected well past the timeout, audio uninterrupted.
2. Playback stopped, still no input -- disconnects roughly one timeout period
   after audio stopped, not immediately.
3. No audio and no input -- disconnects at the timeout, unchanged from
   before.
4. Mic mute still independently suppresses idle disconnect.