Alignment in fast_ofstream

[Cody-G]

It seems that the PCO Edge 4.2 sensor size is less friendly to "4k" hard drive alignment than the PCO Edge 5.5. At the full chip size there is no problem, but when one reduces the vertical size of the ROI then our current fast_ofstream implementation fails.

[Cody-G]

I suspect that my fix will negatively affect write speed, so I'm leaving this issue open until we have a better solution.

[Cody-G]

I think I have a better solution for this, waiting for PCO to fix an (unrelated) problem before I merge

[Cody-G]

I was correct that we lost a lot of write speed with my previous fix. #30 includes a better solution: we write only 4K-divisible chunks so that we stay compatible with the NO_BUFFERING file flag. When image size is not divisible by 4K due to the ROI settings, then we still have a problem when writing the very last chunk of data because it's not necessairly divisible. I solved this by writing a few bytes of extra garbage at the end to keep it divisible. Note that this will cause Julia to complain when loading the image, because the file size is longer than expected. Probably we can improve this in a future PR by shrinking/truncating the file to remove those extra bytes after the recording is finished.

[timholy]

Certainly fine to fix this on the Julia end, but if it's not hard to trunc the file then that does seem better. Might reduce confusion for folks who open files in other programming languages/applications.

[Cody-G]

Yeah it should be just a one or two line change. I will go ahead and do it before we merge this.

[Cody-G]

It took more than one or two lines but I've fixed this in my updated pull request :)