Cut the last release before the switch over to end-to-end ByteString. (Including fixing the rpm repo's index which had not been getting updated.)

I had left the bs branch last week with a known bug, so got that fixed. Also there were some encoding problems on windows with the ByteString filepaths, which needed a new release of filepath-bytestring to clean up. Now I think the bs branch is really in a mergeable state. (It's still not tested on Windows at all though.)

Took the last little while to do some more profiling. Mostly the remaining ByteString conversions barely seem worth doing (1% improvement at most), but ?optimise journal access seems like it could pay off well.

Also found time in there somewhere to implement git annex inprogress --key