fix(mediorum): stop terminal transcode retries

[RolfAris]

Problem

Some bad audio uploads are retried indefinitely by the missed-transcode backlog.

The retry loop is:

Step	Before
select missed jobs	status IN ('new', 'error'), error_count <= 5
queue job	write busy upload update
ffmpeg fails	write error upload update
next scan	same upload is eligible again

That keeps producing full-row uploads/update CRUD ops for uploads that already failed repeatedly and do not have a 320 transcode result.

Change

Stop selecting terminal transcode failures once error_count >= 5 and there is still no 320 result.

Also guard transcode() itself before writing the busy state, so an over-limit upload cannot produce another transient write if it reaches the worker through another path.

Audio-analysis backlog selection now skips the same terminal transcode failures unless a 320 result exists.

Evidence

The source cap is necessary but not sufficient by itself.

In the canary, val008 ran the source-side retry cap without the receiver-side suppression. Its uploads/update growth stayed essentially flat versus its own pre baseline:

Node	Role	pre rows	source-only rows	pre bytes	source-only bytes
val008	control: source-only	4,196	4,109	35,694,662	35,699,538

That is -2.07% rows and +0.01% bytes in the latest 1h sample. Across 20 clean hourly samples, source-only averaged -0.47% rows and -0.53% bytes, so it did not materially reduce validator ops growth.

This PR still matters because upgraded nodes should stop producing local terminal retry churn. The material validator-wide reduction requires pairing this with receiver-side suppression of legacy remote retry ops.

Tests

go test ./pkg/mediorum/server -run 'TestTranscodeRetryLimit|TestFindMissedJobCandidates|TestFindMissedJobsMarksUploadsBusyBeforeEnqueue|TestFindMissedAudioAnalysisCandidates' -count=1