[dawn][vulkan] Removes queue serial assertion.

- The assertion is currently causing flakes after the changes
  in https://dawn-review.googlesource.com/c/dawn/+/242714. The
  change in that CL is necessary to implement spontaneous
  callbacks, but by moving the `CheckPassedSerial` call which is
  the call responsible for actually updating the `mCompletedSerial`
  member in `ExecutionQueue` outside of the scope of the global
  Device lock, the assertion can fail when:

  Thread A:
    - Successfully waited for serial 3 in the critical section of
      `WaitQueueSerials`.
    - Calls `CheckPassedSerial`->`CheckAndUpdateCompletedSerial`
      and gets 3. Note that in the Vulkan implementation of
      `CheckAndUpdateCompletedSerial`, this updates the internal
      fences that are tracked, so all serials < 3 are now no
      longer in `mFencesInFlight`. Begins to try calling
      `UpdateCompletedSerial(3)` but is interrupted by Thread B.
  Thread B:
    - `WaitForQueueSerial` is called for a serial of 2. Since
      the backend implementation, i.e. Vulkan's
      `CheckAndUpdateCompletedSerial` has already been called,
      in `QueueVk::CheckAndUpdateCompletedSerial` we don't
      find a wait fence, so we enter the code block with the
      DAWN_ASSERT. Unfortunately, since Thread A did not
      complete `UpdateCompletedSerial(3)` yet, the actual
      value of `mCompletedSerial == 1`. As a result, the
      assertion fails.
- This temporary change should fix the crashes we are seeing in
  the bugs, but a follow up more extensive change will be done
  to make sure that serials in `ExecutionQueue` are a bit more
  consistent in a multithreaded environment.

Bug: 419598089, 420681969
Change-Id: I8c3002379cec1ce0ab10e314cf3cd9d3a3512314
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/244035
Commit-Queue: Kai Ninomiya <kainino@chromium.org>
Auto-Submit: Loko Kung <lokokung@google.com>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>