Global delay bug fix: check again credits under mutex

[Yuval-Ariel]

i encountered a bug while running tests for the dirty memory connection to the global write controller and the bug is as follows:
during WriteController::GetDelay , we check if there are enough credits and return if there are, here:

auto credits = credit_in_bytes_.load();
  while (credits >= num_bytes) {
    if (credit_in_bytes_.compare_exchange_weak(credits, credits - num_bytes))
      return 0;
 }

if there arent enough credits, we proceed under the mutex but dont check again the condition above which might have changed from a different thread.
this PR adds this additional check and while at it, rename and reorder some members.

[ayulas]

the use of credit_in_bytes_ is not right as i can see it... as you wrote above
"// This function is no longer under db mutex since credit_in_bytes_ is atomic"
in the first check you does threat it as atomic and do compare_exchange_weak meaning you do change its value not under mutex

but after you take the lock , you changed its value not using atomic operation which is problematic
the lock actually doesnt protect from you from access credit_in_bytes_...
you need to access credit_in_bytes_ with atomic functions

[Yuval-Ariel]

right, we'll simply take the mutex before to include the first place where credit_in_bytes_ can change so now only 1 thread can change credit_in_bytes_ simultaneously