Calculate Max N for all classes for multiple datasets

[subdavis]

Something to note: In order to make this fast, the actual detections (features) are not parsed, and instead the track bounds are used. That means that a track that spans 10 frames but that has no detection on a given frame will be counted in the summary for that frame. I believe this is correct from a stasticical perspective: the target is occluded, but exists in the frame.

If this is an incorrect way to count, the summary generator will need to be moved into a worker process and this will add some complexity.

I want to make the summary button it's own dialog with other report types, but this was the quick version. I added some icons to the button to indicate which ones had "submenus" but I could revert this if you want. Until now, all those buttons were "safe" -- they didn't do anything until you confirmed something in another menu. This button immediately generates a report, which could make the user hesitant to explore buttons.

Fixes #710

[mattdawkins]

I believe this is correct from a stasticical perspective: the target is occluded, but exists in the frame.
I want to make the summary button it's own dialog with other report types, but this was the quick version.

+1

Didn't mentally parse the part about buttons but any quick first pass LGTM

[BryonLewis]

I think just the one question about structure of the call to viame_summary, it's not vital and you can use your judgement as to if you want to move it into summary.service.ts

I went through and tested it on 4-5 copies of JRS8/7 and seemed really quick.
I used a smaller SeaLion dataset and verified the numbers and detection counts using the detection timeline viewer.