Maybe we could implement our own wg_format_from_caps_audio() rather than using gst_audio_info_from_caps() directly? Like the other wg_format_from_caps_*() functions, we warn when find missing fileds instead of error.
I don't think that solves the problem. If we're missing the sample rate, that's kind of a serious problem; we kind of need the sample rate in order to report it downstream. Does the stream ever give more specific caps?