qemu does correctly parse the given options when adding a new block device or on
startup (i.e. during bdrv_open). However, after a migration is completed, the
qcow2 layer is reopened in order to refresh stale metadata, and on this occasion
we're losing most driver-specific options.

Most of the options are hard to directly observe, so I used gdb to verify this
behaviour.

$ gdb --args x86_64-softmmu/qemu-system-x86_64 -drive file=/tmp/test.qcow2,overlap-check=none
(gdb) run
<press ctrl-c>
(gdb) p ({struct BDRVQcow2State}bdrv_states .tqh_first .opaque).overlap_check 
$1 = 0

So we get no overlap checks indeed. Start up another instance with the same
options, but additionally with -incoming. You can do the same check, and it
will return 0 again.

Now migrate to the second instance and after migration has completed, press
ctrl-c and do the check another time:

(gdb) p ({struct BDRVQcow2State}bdrv_states .tqh_first .opaque).overlap_check
$2 = 127

This is wrong, it should still be 0 instead of the default value of 127.


Options affected in qcow2 are (from source inspection):
* pass-discard-request
* pass-discard-snapshot
* pass-discard-other
* overlap-check
* overlap-check.main-header
* overlap-check.active-l1
* overlap-check.active-l2
* overlap-check.refcount-table
* overlap-check.refcount-block
* overlap-check.snapshot-table
* overlap-check.inactive-l1
* overlap-check.inactive-l2