scsi_lib: Some love to scsi_lib

Submitted a much needed cleanup to scsi_lib's command completion path. The code after these patches should theoretically be behaviour equivalent to current code. Only cleaner smaller, and easier to understand. The motivation was to make things clearer to understand. It was all inspired by Alan Stern and more things added to that.

Theoretically, since I have only tested with the usual tests I have. Boot an x86_64 machine, git clone, make, xfstests ...
I did not run any Error injection tests with scsi_debug, since I don't have any set up. Though one should because the error handling might have changed accidentally, though not intended.

Please someone, Alan Jeff ... I have tried to go through every possible code path and prove that we do exactly as before. But I might have missed something. Please check my Math.

I'm sitting on these patches since 2.6.34-rc4, but never followed them through, and tested them. Sorry. They patched without any issues on top of current scsi-misc meaning nothing changed in this area since a long time.

The patches are:
[PATCH 1/3] scsi_lib: request_queue is only needed inside scsi_requeue_command [PATCH 2/3] scsi_lib: Remove that __scsi_release_buffers contraption

These two are pure and trivial

[PATCH 3/3] scsi_lib: Collapse scsi_end_request into only user

This one should be reviewed, but should not change any behaviour.

Thanks
Boaz