ntco::IoRingSubmissionQueue: optimize the way 'push' works

[smtrfnv]

Is your feature request related to a problem? Please describe.
When submission queue becomes full the driver automatically enters the ioring (ntco::IoRingUtil::enter) and waits for numToSubmit events to be submitted. My concern is that it submits exactly 1 entry less than it could.

Let's consider ntco_ioring.t TC1. See the log:
ntco_ioring.t_assertion.txt
Submission queue size is 4 entries. From the begining of the TC, entries are continuosly pushed (test->defer(id);) to the driver: ID0, 1, 2.....N.
When pushing item with ID 3, there are already 3 entries (ID0, ID1, ID2) in the submission queue. But it detects that it's full and enters the ioring (ntco::IoRingUtil::enter) with numToSubmit equals 3. From my perspective, that's a missed opportunity to submit all 4 entries.

Describe the solution you'd like
With respect to the example above, I think that IoRingSubmissionQueue::push should work differently:

1. It should check if the submission queue is already full and enter the ioring (ntco::IoRingUtil::enter) if so, pushing all the items from the queue.
2. Save new entry onto the queue.