Add a `reproduce` command

It would be nice if we had a single command to try to reproduce MUSE2 output files for given input files. We could repurpose some of the machinery we have for regression tests to provide details about which files differ and where. As with regression tests, there should be some tolerance to small differences in floating-point numbers (this should probably be configurable, too).

I'm not totally wedded to this being a separate command. We could add a `--reproduce` flag to the `run` command or something instead.

Here are a few applications of this feature I can think of:

- For researchers: Trying to reproduce other researchers' results. (If we make it easy enough, surely everyone will do it... right? 😄)
- For developers: Helpful for debugging a model that's not working for someone
- Checking if/how results change between different versions of MUSE2 (obvs this won't be resistant to breaking changes in file formats, but hopefully these will end up being less common in future)

Other issues that might be relevant to this:

- [x] #1109
- [x] #1186