Pillow has trouble reading certain JPEG2000 files [1], so fall back to
OpenJPEG's opj_decompress when that happens.

It looks like the next release of Pillow will contain a fix, so then we
can finally get rid of these workarounds.

[1] https://github.com/python-pillow/Pillow/issues/3556

This reverts commit 50d0ab19.

Hank requested that we ignore scandata when it is derived, because they
might be out of date

This will prevent the Abbyy converter from emitting empty lines in
special cases.

This brings extra validation of the _meta.xml XML files that we write to
disk. Very rarely we see _meta.xml files becoming corrupted, so perhaps
this will help.

Page timeouts are fatal unless adaptive_ocr=true.
ocr_degraded will be set if a page timeout occurs
Timeouts can be specified as task argument, using 'ocr-page-timeout'.