Audio transcription sent to undeclared/test Google Account and not to the provided llm client

Transcribing audio content will use an unspecified Google API key for the transcription, and not as expected use the provided llm_client.

This is in part solved by #326 which at least provides an option to not route everything this way.

Instead of relying on the provided LLM llm_client markitdown will process audio via the SpeechRecognition library sr:
https://github.com/microsoft/markitdown/blob/da7bcea527ed04cf6027cc8ece1e1aad9e08a9a1/packages/markitdown/src/markitdown/converters/_transcribe_audio.py#L45-L49

In SpeechRecognition recognize_google is mapped to google_legacy:

https://github.com/Uberi/speech_recognition/blob/46e70560f605ed190b3b0c16f198ee34978de585/speech_recognition/__init__.py#L1288

The google_legacy method even comes with this warning (although does not declare where this key comes from and how the data may be used):

The Google Speech Recognition API key is specified by key. If not specified, it uses a generic key that works out of the box. This should generally be used for personal or testing purposes only, as it may be revoked by Google at any time.

https://github.com/Uberi/speech_recognition/blob/46e70560f605ed190b3b0c16f198ee34978de585/speech_recognition/recognizers/google.py#L225-L262

As it's using some unspecified API key
https://github.com/Uberi/speech_recognition/blob/46e70560f605ed190b3b0c16f198ee34978de585/speech_recognition/recognizers/google.py#L118-L119

Start in packages/markitdown/src/markitdown/converters/_transcribe_audio.py at the audio transcription call, then read SpeechRecognition's recognize_google and google_legacy implementations linked in the issue. Determine how transcription should use the provided llm_client instead of an unspecified Google API key, and verify that the resulting path no longer sends audio through that undeclared account.