analysis: don't keep re-analyzing a track whose file is missing

[vynulldev]

What & why

The web UI asks for a track's waveform each time it renders the row, and for a track with no analysis yet every request queued one. When the track's file is missing (an imported path that doesn't exist on this machine), each of those analyses ran ffmpeg and failed the same way. A CDJ test session with one missing track logged 14 failed analyses in two minutes, and it would carry on for as long as the row stayed on screen.

• Queueing an analysis now checks that the file exists. A missing file is skipped, and the track is flagged as missing in the library, which the web UI shows.
• The check runs on every request, so a file that comes back (restored, or relinked) is analyzed on the next one.
• The lazy analysis a deck's request triggers skips a missing file too, instead of starting ffmpeg.
• So does on-demand analysis, which the web UI's track drawer triggers for its waveform and beat grid: opening a missing track used to run three failing analyses at once.
• Tracks that already have analysis are unaffected: a cached or imported analysis is still served when the file is gone.

New tests check that repeated requests for a missing file queue nothing and flag the track, that the track is queued once its file is back, and that on-demand analysis skips it; they fail without the fix. The existing queue test now uses real temporary files, since it relied on paths that don't exist. Conformance against rekordbox 7.2.19 is unchanged (3,054 of 4,250 exact, none gained or lost).

Hardware testing

• Tested on: CDJ-2000NXS2

Checklist

• go build ./..., go vet ./..., and go test ./... pass
• gofmt -l . is clean
• New source files carry an SPDX header (GPL-3.0-or-later)
• Tested on real hardware (deck + firmware noted above), or this change doesn't affect deck behaviour
• I agree my contribution is licensed under the project's GPLv3