record_linux: parecord exit is not handled — no error reason, state stays record, stop while paused may leave it running #638

Package version

record 7.1.2, record_linux 2.1.3

Environment

• OS: Linux (Ubuntu 24.04, PulseAudio)
• OS: Linux (Ubuntu 22.04, PulseAudio)

not tested yet, but will do later during fixing

• OS: Linux (Ubuntu 26.04, PulseAudio)

Describe the bug

CapturePipeline starts parecord but never looks at how it ends:

1. No reason when it fails.
   • stderr is drained into nothing and the exit code is never read (capture_pipeline.dart:98-106).
   • If parecord cannot connect (unknown device, no sound server), it exits with code 1 and prints the reason,
     for example Stream error: No such entity.
   • startStream() still succeeds, and the stream simply ends with 0 bytes. The caller cannot tell a failed start
     from a normal stop.
2. The state stays record. When parecord ends on its own (failed start, crash, device gone), isRecording()
   stays true and onStateChanged() emits no stop. Only stop() resets the state (linux_recorder.dart:72-82).
3. Stop while paused.
   • pause() freezes parecord with SIGSTOP (capture_pipeline.dart:72).
   • stop() then sends SIGTERM and drops the process (:81-82), without SIGCONT.
   • A stopped process does not act on SIGTERM until it is continued, so parecord can stay alive and keep the
     microphone open.
   • In general, stop() does not wait for the exit and has no SIGKILL fallback.

RecordConfig(encoder: AudioEncoder.pcm16bits, sampleRate: 16000, numChannels: 1)

To Reproduce

Steps to reproduce the behavior:

1. startStream() with device: InputDevice(id: 'no-such-source', label: 'x').
2. The stream ends at once with 0 bytes and no error event; isRecording() returns true.
3. Start a stream, pause(), stop(); then pgrep -a parecord (item 3).

Expected behavior

• An unrequested exit of parecord (non-zero code or signal) puts an error on the stream before done, with the
  exit code and the last stderr lines. A requested stop() still ends with done only.
• The state goes back to stop when parecord ends.
• stop() ends parecord for sure: continue it if it is paused, wait for the exit, and kill it if it does not end.

Additional context

• Follows AudioRecorder: the stream from startStream never ends when the platform's capture ends #632 (the stream now ends when capture ends; this adds the reason and the state) and record_linux: every AudioRecorder shares one parecord process, so starting or disposing any recorder kills another's capture #631.
• fix(linux): drain ffmpeg/parecord output so encoding never stalls mid-take #626 is why stderr is drained today; keeping a bounded tail keeps that fix.
• Open PR feat(linux): capture with ffmpeg/ALSA, for machines without PulseAudio #629 (ffmpeg/ALSA backend) touches the same file; the fix can apply to both capture programs.

I would like to send a PR with tests for this.