Fix ffmpeg redirects. Fix max resolution setting.

[asandhu3]

Fixes #123.

Changes

• FFmpeg/ffprobe input hardening (probe.js, remux.js, transcode.js, transcodeSession.js)

  • -http_persistent 0 — forces a fresh connection per hop so https → http redirects resolve cleanly. Was reading 302 body as stream content.
  • -reconnect_on_http_error 4xx,5xx — handles intermittent 407 Proxy Authentication Required from the CF edge (FFmpeg only special-cases 400/401/403/404 natively).
  • -reconnect_delay_max bumped from 3/5 → 10 to cover CF's rate-limit window.

• Max Resolution now actually caps (transcodeSession.js — buildScaleFilter)

  • Previously scaled to the exact target height regardless of source resolution, so Max = 4K upscaled 1080p sources to 4K. Now computes Math.min(targetHeight, sourceHeight) in JS and emits a literal integer
    into the filter (scale_cuda=-2:1080:... instead of scale_cuda=-2:min(2160,ih):...). Testing showed the expression form triggers a degraded scale_cuda code path — literal 1080 ran at 1.8x realtime vs 0.69x
    with the expression, on identical source.
  • videoHeight plumbed through: probe → POST /api/transcode/session → TranscodeSession.options → scale filter.
  • Graceful fallback when source height is unknown (probe skipped/failed): uses the configured target directly, same behavior as before the cap feature.

• Negative-cache probe failures (probe.js)

  • Failed probes now populate the cache with a shorter effective TTL (~30 s) so a client retry doesn't immediately re-hit the provider and double the connection count on rate-limited accounts.

• Frontend: VideoPlayer.js and WatchPage.js pass videoHeight: info.height from the probe result into the transcode session POST body.