Add local Whisper transcription and transcript UX fixes

[Fiona-neena]

Summary

This PR improves local development for writeout.ai and tightens up a few transcript UX issues.

What changed

• added a configurable transcription driver flow
• added local Whisper transcription support for real local audio transcription
• kept translation configurable and safe for local development
• added local mock transcript support for development fallback
• allowed guest uploads in local mode when enabled by config
• made transcript storage use the configured disk consistently
• improved the 403 page CTA for guest/local flows
• fixed the transcript player mount markup
• cleaned up layout encoding issues in the shared Blade layout
• changed transcript download to export clean plain text instead of raw VTT timestamps
• updated local setup docs and env examples
• ignored Python cache artifacts in git

Local setup notes

• local transcription now supports WRITEOUT_TRANSCRIPTION_DRIVER=local_whisper
• local translation can stay on WRITEOUT_TRANSLATION_DRIVER=mock
• Hugging Face cache is kept inside storage/app/huggingface
• local Whisper is configured to use the CPU-friendly model by env

Why

The main goal was to make local development usable without OpenAI billing, while still allowing real transcription on a local machine and improving the transcript experience in the app.

Testing

• verified PHP syntax on changed PHP files
• verified the local Whisper Python script compiles
• installed and smoke-tested local Whisper on Windows
• verified Laravel can launch the Python transcription process
• verified transcript downloads now produce plain text output