Most virtio_blk requests are launched from vtblk_startio; prior to this
commit, if vtblk_request_execute failed (e.g. due to a lack of space on
the virtio queue) vtblk_startio would requeue the request to be
reattempted later.

Add a flag "vbr_requeue_on_error" to requests and perform the requeuing
from inside vtblk_request_execute instead.

No functional change intended.