Add private RunPod generation provider

[SohailAwg]

Summary

Adds a third video generation provider: Private RunPod API. This lets LTX Desktop run generation on a user-owned RunPod Pod instead of the local desktop GPU or the official LTX API.

Key changes:

• Add RunPod provider settings and UI copy alongside local and official LTX API modes.
• Add a private /v1/* FastAPI compatibility layer for text-to-video, image-to-video, audio-to-video, retake, uploads, and prompt enhancement.
• Add RunPod container files and setup docs.
• Support LTX 2.3 Fast model variations:
  • fast = Distilled 1.1
  • fast_legacy = Distilled 1.0
• Preflight selected model downloads on the Pod before generation to avoid first-use RunPod HTTP proxy timeouts.
• Fix private upload URLs so desktop clients do not try to PUT media to internal Pod IPs.
• Keep official LTX API and local GPU generation support intact.

Validation

Ran locally on Windows:

• pnpm backend:test — 263 passed, 1 warning
• pnpm typecheck:ts — passed
• pnpm build:frontend — passed, existing chunk-size warning only

RunPod coverage included in tests:

• RunPod provider routing for T2V and retake
• private upload URL rewrite
• model download preflight polling
• RunPod/local model spec validation
• prompt enhancer flag propagation

Live RunPod note:

• This branch was previously exercised against an A100 80 GB RunPod Pod with 720p I2V through the desktop backend and prompt enhancement enabled.
• I could not rerun a live RunPod smoke test while opening this PR because the test Pod now returns pod not found from the RunPod control API and the account currently lists no active Pods.

Privacy / behavior

Private RunPod mode sends prompts and media only to the configured user-owned RunPod URL. Official LTX API mode remains separate and unchanged.