feat: large recording playback with chunked streaming

[fank]

Summary

This PR implements chunked streaming playback for large recordings that would otherwise crash browsers due to memory constraints. Recordings are converted from JSON to binary formats (Protobuf/FlatBuffers) and split into chunks that are loaded on-demand during playback.

Key Features

• Chunked binary formats: Convert large JSON recordings to Protobuf or FlatBuffers
• On-demand streaming: Load only the chunks needed for current playback position
• Event-driven conversion: Conversion starts immediately after upload (no waiting for cron)
• Background worker: Retries failed conversions and handles batch processing
• Browser LRU cache: Keep 3 chunks in memory for smooth playback
• Optional persistent cache: Enable with ?cache=1 URL parameter

Architecture

Upload → Save JSON.gz → Trigger Conversion → Create Chunks
                              ↓
                    manifest.pb (metadata, entities, events)
                    chunks/0000.pb, 0001.pb, ... (frame data)
                              ↓
Playback → Load manifest → Stream chunks on-demand

New Endpoints

Endpoint	Description
GET /api/v1/operations/:id/format	Get storage format info
GET /api/v1/operations/:id/manifest	Stream manifest (binary)
GET /api/v1/operations/:id/chunk/:index	Stream chunk (binary)

Configuration

{
  "conversion": {
    "enabled": true,
    "interval": "5m",
    "storageEngine": "protobuf"
  }
}

Files Changed

• Backend: Storage engine interface, Protobuf/FlatBuffers engines, conversion worker
• Frontend: ChunkManager, StorageManager, streaming playback in ocap.js
• Schemas: Protobuf and FlatBuffers definitions

Test plan

• Upload a small recording, verify immediate conversion
• Upload a large recording (500MB+), verify chunked playback works
• Verify playback controls (play, pause, seek) work in streaming mode
• Verify legacy JSON recordings still work
• Test with ?cache=1 to enable browser caching