scsi_debug: add support to corrupt data and/or reftag v3

[Christoph Hellwig <hch@lst.de>]

This will be used for testing PI support in xfstests.

Sashiko complains about a few existing issues, and "duplicate"
existing issues because of the capacity wraparound.  I do not
plan to address them as the use case of wrap-around does not
overlap with this, and fixing it partially would be a mess.

Similarly the long non-preemptible sections if injecting a long
consecutive range of LBA are intentional, but probably a bad idea
for a user to perform.  But putting an abritrary limit on the
number of blocks would not be helpful either.

Changes since v2:
 - protect against fake_rw=1,

Changes since v1:
 - use match_int to parse the  reftag_adjust option
 - don't bypass error cleanup on parsing failure
 - allow to corrupt the entire capacity
 - set the bug_if_fake_rw to devip2sip to false
 - fix the check for unmapped LBAs
 - require DIX for reftag corruption injection
 - make the kernel messages a bit less verbose
 - add calls to sdeb_data_lock and sdeb_data_sector_lock