fix(orchestrator): ensure input tempfile cleanup on all exit paths

[chengyixu]

Fixes #1716

/claim #1716

Problem

When Transcode() returns an UnrecoverableError, transcodeSeg panics (line 749). This panic bypasses the explicit os.Remove(fname) call at line 799, leaving *.tempfile input files on disk. Over time, this fills up the orchestrator's data directory.

The terr() error helper also cleaned up the file, but it is never reached on the panic path.

Solution

Move the input tempfile cleanup into a deferred function at the top of transcodeSeg so it executes on all exit paths:

• Normal return (successful transcode)
• Error return via terr()
• Panic from UnrecoverableError

The existing keepInput flag (for preserving input when extremely large outputs are detected) is hoisted to function scope so the defer can check it.

Changes

• core/orchestrator.go: Add defer cleanup for fnamep; remove redundant cleanup from terr() and the explicit os.Remove block
• core/livepeernode_test.go: Add two regression tests:
  • TestTranscodeSegCleansTempfileOnUnrecoverableError — triggers panic path, asserts no leftover tempfile
  • TestTranscodeSegCleansTempfileOnNormalError — triggers normal error path, asserts no leftover tempfile

Testing

• Added regression tests covering both panic and normal error cleanup paths
• Existing TestTranscodeAndBroadcast continues to pass (deferred cleanup handles the normal path)
• CI workflows should validate full test suite

[j0sh]

We want to keep the tempfile on error for later inspection.