BUG: Parallelization over runs randomizes report order

[larsoner]

Currently we parallelize over subjects and runs. This leads to stuff where in parallel mode run 4 can be processed faster than run 1, and thus appears in report first. In serial mode this won't happen and run order is preserved. I've opened mne-tools/mne-python#12424 to tackle this, but once that (or something similar) lands, we should also reconsider ordering more generally. Take for example this TOC:

It would be better to ravel the sections differently, namely have all run 1 parts first, then all run 2 parts second, etc. Typically a user wants to compare these bits sequentially, and I find it annoying to have to scroll super far to do it.