always accurate progress bar

[nothings]

The current code has the option to count all the faces first, then to use that to make an accurate progress bar. But it's not recommended for large numbers of files. Is that recommendation because it has to find the faces twice, which is a performance hit, or because it has to run for a long time finding the faces without giving any feedback on the progress bar?

I'm thinking that for both of those issues, we can potentially address them.

For the first one, the overhead of finding faces twice, we can count the faces first, but save the face information we computed in the process, and then on the second pass, reuse the saved information instead of recomputing it. That way it will go no slower than the non-CountFaces way. I absolutely know what to do to make this work.

For the second one, I assume the progress bar normally only gets updated by the Automatic1111 framework when an SD job is issued, so we'd have to find a way to update it in the interim, maybe there's an explicit way to do it, or maybe we can do a hack where we generate a dummy SD job that has 0 steps or something so it doesn't do any work. Then generate fake progress updates as we scan the files counting the faces. Maybe estimate that face counting takes 5% of the total time. It won't be accurate progress measurement until we start the actual processing, but it would show something.