Loads all transcription data from the analysis folder into memory.
This enables the /search/transcript endpoint. Transcription data is also
loaded automatically when using /analysis-data/load.
Does not require a specific whisper model to be loaded — transcriptions from
any supported model are loaded.