MediaRiffle v0.8.0

✨ Features and improvements

• Full screen on a phone: a ⌄ button hides the controls without pausing; with them hidden, a tap brings them
  back instead of also pausing the film; and they fade by themselves again after a tap (the last control tapped stayed
  "hovered" on a touch screen, which kept them up). The 🚫 button that hides a video is just its icon now, as
  "Hide" written out was taken for hiding the controls.
• Blu-ray (PGS) subtitles without converting the video: they're drawn by the browser over the picture (libpgs),
  so turning them on or off never restarts the stream and a 4K file keeps playing as it is. Subtitle tracks are
  extracted ahead of time in the background (all of a file's in one read, newest files first, only while nothing is
  streaming) and kept, up to SubtitleCacheGB, so turning one on is instant; text subtitles of a big file no longer
  take minutes the first time either. DVD subtitles, and any picture subtitles on a TV, are still burned in.
  PrepareSubtitles turns the preparing off.
• A 1440p level, between 1080p and 4K: "MaxHeight": 1440 (with "MaxVideoKbps": 16000) offers it, 2160 both.
  For a GPU that can't convert 4K in real time but can to 1440p (an Intel Arc A380: 1.2x from a 60 Mbit/s 4K HDR
  disc rip, against 0.6x to 4K); Auto only steps up to it while the encoder keeps ahead.
• Changing stream without stopping: picking a quality, an audio track or an image subtitle, and Auto moving
  between the file as it is and the converted stream, used to stop playback and show Loading… while a new stream
  started. Now the new stream starts in the background, a few seconds ahead, while the old one keeps playing, and
  takes over once it has the next few seconds ready: no pause. (Paused, casting, AirPlay and the last seconds of a
  file change the plain way.)
• Bring back lists, favorites and watch history after a move: deploy/export-user-data.sql (run on the old
  database, in DB Browser for SQLite or sqlite3) makes an import script for the new one; deploy/carry-user-data.sql
  does it in one step with both databases at hand. They carry them from an
  old database into one whose libraries were scanned afresh (moved to a new server, or a library's folder changed in
  Admin), matching shows, movies, music and files by what they are rather than by id. See docs/proxmox.md.

🐞 Bug fixes

• Carry user data (!24) (@acalcutt)
• docs/proxmox.md: moving an install in, upgrading, HAProxy, as tippr's (!23) (@acalcutt)
• The end of a film is the end: a stream that fails in the last 20 s (or last 1% of a long film), as a file's
  stated length can run past what there is to play, now counts as finished (saved as watched, on to the next
  episode) instead of a playback error, or a new stream started at the very end. It finishes once, too, so a play
  queue never skips one.
• Quick Sync streams kept their segments and clock in step: Intel's encoder ignored the forced keyframes every 4 s
  without -forced_idr, keeping its own 256-frame GOP, so segments came out 10 or 21 s long against a playlist of 4 s
  ones. The stream's clock drifted from the file's, and near the end a film could run on past its length or stop with
  a playback error.
• Play works after a long pause: a video paused overnight (the computer asleep, so the player's "still here" never
  reached the server, which stopped the stream as idle) couldn't be started again without reloading the page. Play
  after a pause of over a minute now checks the stream is still there and, if not, starts a new one from the same
  place. A stream that fails while paused is replaced at the next Play rather than at once (where it would go idle
  too), recovery re-arms instead of giving up after one try, and the file played as it is recovers too.
• Auto stops returning to a 4K original that stalls: once the file as it is stalls on a device (a 60 Mbit/s
  disc rip, say), Auto remembers it there for 30 days, starting on the adaptive stream and not stepping back up to
  the original every two minutes, which made playback stall and switch quality over and over. Original from the
  menu still plays it.
• 4K HDR on an Intel Arc (VA-API) converts at 1.7x real time: the GPU now decodes and scales it, and only the
  1080p frames are copied back for libplacebo and encoded on the GPU again. Handing VA-API's 10-bit frames straight to
  Vulkan crashed on Arc with Ubuntu 24.04's Mesa, so HDR fell back to the CPU and couldn't keep up. The startup
  libplacebo check waits long enough for a first run's shader compile (it gave up after 30 s), and with VA-API, HDR
  decoded on the CPU is tone-mapped with libplacebo too rather than zscale.