https://github.com/w3c/mediacapture-record/issues/119 covers a lot of why this is difficult. Having a live recording API that can request data on demand makes it difficult to produce seekable files because at the time you're writing the index for seeking you don't yet know all the seek points. My recollection was this made it effectively impossible to do this in webm without a remux (this topic comes up in the linked github issue).

It may be possible to do in containers that support offsets from the end of the file such that the header could contain info along the lines of 'the index will appear just before the file ends, so check there', and we can then write the index at the end of the file.