fix: treat PlaceMarker failure as non-fatal and shut down sink on error

[scream870102]

Summary

Fixes #156.

On recent Windows 11 builds (observed on OS build 10.0.26200 with mfmp4srcsnk.dll 10.0.26100.8457), IMFStreamSink::PlaceMarker on the OS-provided MPEG4 media sink returns MF_E_INVALIDTYPE for every marker type and argument combination. The ? on the PlaceMarker call killed the stream event loop, the finish signal was never sent, and CompleteAsync failed with Failed to complete muxer: channel closed (a hard Rust panic on versions before better_error_handling).

Changes

• PlaceMarker(ENDOFSEGMENT) failure is now logged and treated as non-fatal. The marker is only used as a completion hint; finalization still runs and writes the moov box without it.
• The muxer task now calls sink.Shutdown() on the error path, so the output file handle is no longer leaked until process exit when muxing fails.

Verification

• On an affected machine (build 10.0.26200), export previously failed 100% of the time; with this change, StopAndExportAsync completes and produces a valid, playable mp4 (moov present). Confirmed on two machines with the same build.
• The standalone C++ repro from Windows: export always fails with "Failed to complete muxer: channel closed" — OS MPEG4 sink rejects PlaceMarker (MF_E_INVALIDTYPE) #156 confirms PlaceMarker is rejected at the OS level independently of this package, so skipping it is the only client-side mitigation available.

Native binaries are intentionally not included; they can be rebuilt via the Build unienc workflow.

🤖 Generated with Claude Code

[ruccho]

Thank you for the report. This indeed seems to be specific to that build... I could not reproduce it with mfmp4srcsnk.dll (10.0.26100.8655) on my Windows 11 (26200.8655) (all PlaceMarker calls succeeded in mp4sink_repro.cpp). Treating PlaceMarker as non-fatal seems fine, so I will go ahead and merge this.