Revert "Merge branch 'improve_error_capture_step1' into 'xe'"

This reverts merge request !196 (merged)

[6:19 PM] Hellstrom, Thomas Hey Lecluse, Philippe, the new vma error capture code is causing deadlocks apparently since it grabs the vm lock in the dma fence signalling critical path: Since we're allowed to wait for dma_fences under the vm lock, that's not allowed..

[6:40 PM] Hellstrom, Thomas Yes, I think so (if I am correct in assuming that the capture happens before the reset?). One way to work around this that I think is used by i915 VM_BIND is to have a VM list of vmas to be captured on error, and that list is protected by another lock that we don't allow waiting under. Then it's safe to traverse that list in the dma-fence critical path. Daniel's preferred direction WRT this was to do capture after the reset, assuming the engine would be banned so nobody could alter the contents on the VMA. Then we could even do funny stuff like move the contents from unmappable to mappable to access it. Drawback would be that other engines on the same VM could still alter the contents.....