[🐞 BUG] AI post-processing errors typed into user's document instead of falling back to raw transcription

[uid4oe]

When AI post-processing is enabled, if the underlying LLM API call fails, the error message is returned as the "final text" and typed directly into the user's active application.

For example, a user dictating into a document could see Error: HTTP 503: This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later. inserted instead of their spoken words.

Steps to reproduce:

1. Enable AI post-processing in settings
2. Trigger any LLM error (model overloaded, network issue, invalid API key, etc.)
3. Start dictation, speak a sentence, and complete dictation
4. Observe the error string typed into the active app instead of the transcribed text

Affected code paths in ContentView.processTextWithAI():

• Empty API response → returns "<no content>"
• Any LLM error → returns "Error: {description}"

Expected behavior: When AI processing fails, the app should gracefully degrade and return the raw transcription to the user. The error should be logged for debugging but never surfaced as typed text.