A .docx file is a zip containing a one or more .xml files that describe the document. I once had the idea of of giving meaningful names to images in one...after all, it's xml and xml is just text and supposed to be human readable. It isn't. Word documents are rather complex, that's where the "one or more .xml files" comes in. It's why Microsoft built its own open standard back in the aughts when Sun was arguing that OpenDoc was good enough, i.e. Microsoft would have had to remove functionality from Office to implement it.
Which is to give some background on why .docx -> .md -> .docx might not do what users want it to do, if what users want at the end is what they put in at the beginning or something close to it. Markdown simply does not contain as much information as .docx. Even "only" going to .html will probably be a mess: last time I dealt with an html file generated from Word, it was full of local markup <span>'s, but it might have become better since it's been a while.
Anyway, if I were trying to version track .docx files, I'd probably do something like unzip -> commit -> zip and forget about trying to read diffs in the repository.
Good luck.