Interrupted acquisitions

[bryantChhun]

This PR fixes issue #25

The way I solve the problem for singlepage tiffs is different than for multipage tiffs:
multipage tiffs
I use ome metadata, as per Ivan's suggestion, which correctly reflects the final dimensionality

singlepage tiffs
The code already parses the coordinate-to-filename mapping from mm metadata. That mapping reflects actual data --> files that definitely exist in the folder. The code in this PR simply combs through all coordinates and extracts the number of each dimension from there.

I've also included simulated data (stopped early), and appropriate tests for that

[camFoltz]

Do we want to implement acquisitions that fail mid-position, i.e. they only acquire a few channels of that position? And if so, we load the last completed timepoint/position? We would probably have to load the last completed timepoint in this case since it is the outermost dimension. Having timepoints with different dimensions might not be possible or would be a headache for reconstruction

I am open to merging this for now and revisiting later if you think that is best

[bryantChhun]

Do we want to implement acquisitions that fail mid-position, i.e. they only acquire a few channels of that position? And if so, we load the last completed timepoint/position? We would probably have to load the last completed timepoint in this case since it is the outermost dimension. Having timepoints with different dimensions might not be possible or would be a headache for reconstruction

I am open to merging this for now and revisiting later if you think that is best

Interesting point. It's not clear to me. I think cutting out that last frame makes sense from a processing standpoint, but not from a data-purity standpoint (you can easily cut out that last frame manually afterwards, but do we want to leave this to the user?)

If we decide to automatically cut out incomplete "frames", it should be easy enough to add a single function to check "completeness" every time point.

My feeling right now is to leave this for another PR. Another thought -- we should think about the possible space of "incomplete" datasets -- it's not likely to have partial X and Y for example, but for other dims, at what threshold should we cut out the data?

[camFoltz]

Discussed in person. Tests and Changes look good. Will leave mid-position failed acquisition for another PR