Fix transcription and data saving

[becky-gilbert]

This PR fixes the transcription and data saving.

• Adds two new properties to the plugin's data:
  • transcript: the full Whisper transcript text
  • transcript_chunks: word-level timestamps from Whisper
• Adds a module-level _transcriber variable and getTranscriber() function so the Whisper model is loaded once and reused across trials, rather than reloaded on every call.
• Changes the class's data property to be a Blob and removes the arrayBuffer property (the intermediate FileReader/ArrayBuffer step is not needed (array buffer is created in transcribe but it doesn't need to be stored on the class instance)
• transcribe function now takes the audio Blob instead of a URL, decodes, resamples, downmixes, and passes it to transcriber, then writes the transcript/chunks into the trial data passed to finishTrial. Everything is wrapped in try/catch.
• endTrial has been updated to use the new call signature for transcribe
• Removes WaveFile import (no longer used), commented-out code, and hard-coded URLs