Avoid reading h5 imagesets as having a single image, when a format class is not available.

[jbeilstenedmands]

Attempt to fix an issue described in dials/dials#2944

I will try to create a minimal reproducer of the issue, but my basic understanding of the underlying cause is that these parts of the code should not always be defaulting to use Format when the true format class is not known (as we can't pass an image range to set up the 'Reader' correctly). That is appropriate for a e.g. cbf template but not data from a master.h5. Rather we need a FormatMultiImage in these cases. This kind of pattern seems to happen in other functions in this file, and I guess it is not a common code path for imageset creation as we usually know the format.

The dxtbx tests and dials command line tests all seem to pass still.

[dagewa]

The original function is a bit messy - it has specific behaviour according to whether the string "master" is in the template, which is surely fragile, and implies this special case is only intended for *_master.h5-style DECTRIS data. There are other image stack types, like MRC files, and I suppose for those other stack types we'd also want to create FormatMultiImage.

However, I think fixing all that would require rather more invasive changes. I'm satisfied that this is a reasonable fix for dials/dials#2944, and it seems that errors like that are rare in practice.