Er/bundler updates 260112

[earosenberg]

Significant changes and updates to the bundling code. Main changes:

• Atomic db is now required only when creating the bundle db. Coaddition depends on bundle db only; atomic_db is no longer a valid argument. Instead, a new table atomic has been added to the bundle db; this contains columns obs_id wafer freq_channel split_label median_weight_qu basename for every atomic map that exists and passes selection criteria (query_restrict, atomic_list). Maps for coaddition are taken from this table instead of making a new query.
• atomic_list is now applied to the bundle db, so only wafer/obs in the list will be included in the db. This will change median values for inter-obs splits when atomic_list is used. atomic_list can still be passed as an argument to the Coadder to further restrict what is coadded if desired.
• Substantial code simplification to the Coordinator class by using pandas.DataFrame instead of raw numpy arrays. Bundles are still assigned randomly per-obs with number obs/bundle forced to be equal; however the bundle assignment will be different for the same seed relative to old code.
• Because it can no longer be inferred from the atomic_db, map_dir has been changed from an optional to required argument in all functions.
• Docstring fixes, updates, etc

[earosenberg]

Switched back to draft as I want to simplify the Coadder too, using the new improvements to the Coordinator.

[earosenberg]

After testing with an ISO like config this seems to be working for bundling and sign-flip. Have not tested with filtering looking through coadd_filtered_sims.py I don't think it should break.
Given the scale of changes would appreciate a review (eg @kwolz @adrien-laposta @susannaaz) before merging. Any help with further testing would also be welcome.

[kwolz]

Happy to review and test this branch on v4 sim filtering and coadding.