Separate ROI tables computation from image montage

[imagejan]

The way ROI tables are currently computed is far too convoluted:

https://github.com/fmi-faim/faim-hcs/blob/6a1cd2b9d56037f84f523050c99fb347ec009222/src/faim_hcs/MetaSeriesUtils.py#L175-L186

We first determine pixel coordinates for image stitching, then calculate back physical coordinates for the ROI tables. We can instead go directly from the stage position data to the ROI tables (only requiring the shape of a single image once to be converted to physical coordinates).

Also, the montage functions currently violate the single-responsibility principle as they mix pixel stitching/blending with ROI table creation. Separating this will be important to simplify code maintenance, and e.g. for fully supporting dask arrays in the future (see #42).

/cc @tibuch @jluethi

[tibuch]

Addressed in #59.