[feature] refactor screenpipe architecture

[louis030195]

atm:

• CLI
  • core
    • vision (OCR, UI, screenshot, mp4 encoding, DB)
    • audio (VAD, diarization, transcription, mp4 encoding, DB)
  • API server
  • pipes

what we should have:

• API server runs all the time, no matter things are being recorded or not
• pipes runs all the time, no matter things are being recorded or not
• vision and audio clearly separated and can be started / stopped easily, quickly, with smooth shutdown
• things correctly stop themselves when parent is hard killed, or soft stopped

something like

screenpipe
├── api-server (always running)
│   ├── REST endpoints
│   └── WebSocket connections
├── core-services
│   ├── vision-service
│   │   ├── screenshot
│   │   ├── ocr, ui
│   │   └── video encoding
│   └── audio-service
│       ├── vad
│       ├── diarization
│       └── transcription
└── pipe-runner (always running)
    └── plugin management

screenpipe-api    # always running, lightweight
screenpipe-vision # heavy CPU/GPU usage
screenpipe-audio  # heavy CPU/GPU usage
screenpipe-pipes  # sandboxed plugins

maybe it should be different binaries, etc.

maybe we should run a whisper API also?

@EzraEllette @neo773 @ologbonowiwi @tribhuwan-kumar

thoughts?