[Feature] Make `cellfinder` CFOS ready

[alessandrofelder]

Is your feature request related to a problem? Please describe.
Through "natural fluctuations" in @matham and the core developer team's work priorities, focus and workload (and waiting on a MONai release), recent PRs ( #532, #493) tend to hang around for a long time - it's hard to find the contiguous time to get stuck deeply into this and do it justice as a reviewer. Additionally, recent PRs have drifted in scope, which is normal, as we experiment.

This is an attempt to encourage discussion and come up with a plan to effectively include the amazing performance improvements by @matham, which open cellfinder as a viable tool for CFOS and similar data (millions of cells) while maintaining backwards compatibility and current level of ease-of-use for the original main application of cellfinder (<100'000 cells, in few regions).

My instinct is to split up this large goal into smaller steps, and regress against existing test data (both CFOS and a few original serial2p stack?) at each step, checking both performance and consistency of the cellfinder result (within a margin) - I suggest a series of such steps in the order I would do them below, but I am open to objections/alternatives!

Describe the solution you'd like
Overall, I would pursue a strategy of

• hiding extra CFOS parameters (like splitting) in napari
• giving all parameters sensible defaults (usually the value they had for original application?)

Sub-issues 1-3 are preparatory

My understanding of desired features for CFOS suggested in current PRs is reflected in subissues 4-6

Additional context
#464 #517

[adamltyson]

Just my 2c, this is less about imaging modality or number of cells, but more the typical expt/marker, e.g.:

• cFos often LSFM, but not always, often nuclear stains, varying intensity, often high background
• tracing (e.g. retrograde rabies), often whole-cell labels, often overlapping, neurites also labelled