Summary

• route background-task MEDIA outputs by file type instead of sending every extracted media path through the document path
• preserve the voice-media behavior for [[audio_as_voice]] responses so background tasks send audio replies through send_voice()
• add a regression test that exercises a background task response containing a voice MEDIA tag

This issue was partially mitigated on current main because the loop already unpacks (media_path, is_voice), but the background-task delivery path still ignored the media type/voice flag and treated the extracted file as a generic document attachment.

Closes #12064.

Testing

• python3 -m pytest -o addopts= tests/gateway/test_background_command.py