Replace Detected by Transcribe for awake_wav & debug rec

[chatziko]

Currently the awake sound and debug recording are triggered by detection. This PR changes them to be triggered by transcribe for the following reasons:

1. Conceptually it's more accurate, the sounds mark the ASR period during which the user is expected to talk. transcript is already used to mark the end of the period, so the symmetric canonical choice is to use transcribe for the beginning.

2. transcribe is always sent by the server so the code becomes simpler and common for all satellite types, and can be moved to SatelliteBase.

3. Using transcribe prepares the code for implementing a "push to talk" feature, allowing the satellite to be activated by the server without a wake word, going directly to ASR. In this case we'll get transcribe without detected, but we still want the awake sound and debug recording to happen.
   (A separate PR for this feature will be submitted soon.)

Unless I miss something this change shouldn't break any existing setups (since transcribealways comes immediately afterdetection`). I tested a few configurations, it seemed to work fine.

[bazuker]

+1 on this as it also solves a problem of multiple satellites playing awake sound when a wake word is detected by many (considering openwakeword runs on each satellite).
An "error" event is received when wake word collision is detected and thus the "transcribe" event is only received by one selected satellite which makes sense.