drm/i915/gt: Move the late flush_submission in retire to the end

Avoid flushing the submission queue (of others) under the client's
timeline lock, but instead move it to the end so that we may catch more.

References: https://gitlab.freedesktop.org/drm/intel/-/issues/1066
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200420125356.26614-2-chris@chris-wilson.co.uk