[Intel-gfx] BAT bug 95634, take 4

[Chris Wilson]

Since I touched upon an issue with initialisation ordering, trying to
fix an issue between init vs suspend and delayed userspace became a
whole lot more messy with a dip into drm to try and straighten out our
dev->driver->load callback.

There is one major stumbling block in the use of drm_dp_aux's i2c
adapter long before i2c is able to be used (since i2c demands
registration with userspace prior to use!). That's quietly avoided by
initialising enough of drm_dp_aux ourselves to use i2c_transfer, and a
trio of patches to expose the same functionality from i2c await
response. In the meantime, I would like to close the regression...
-Chris