Refactor preprocess_tod to use per-group parallelization

[mmccrackan]

Work in progress. Refactors main preprocess functions to parallelize over each individual detector grouping instead of just over observation id's so as to line up with mapmaking.

To-do:

• Refactor multilayer_preprocess_tod
• Merge both preprocess_tod and multilayer_preprocess_tod's underlying functions with preproc_or_load_group.

[mmccrackan]

Just realized that making the error db an obsdb instead of a manifestdb would make more sense here as it isn't pointing to any files, though since it is keyed by obs and wafer we might need #1159 first. Will see about updating it.

[msilvafe]

Let's close this PR and split up this into two PRs one for the improved error raising and one that implements the per-group parallelization.