D3D11: Implement automatic backend buffer mapping

This CL introduces a new device toggle "auto_map_backend_buffer" that
enables automatic buffer mapping management in the D3D11 backend. When
enabled, the backend maintains buffers in a mapped state whenever
possible, allowing the frontend's MapAsync and Unmap operations to
become lightweight operations that only update buffer state without
acquiring backend locks such as the D3D11 device context lock.

Implementation:
- Backend automatically maps buffers when they are not in use by the GPU
- Before GPU usage, TrackUsage() automatically unmaps buffers
- After GPU operations complete, the queue automatically re-maps buffers
  via scheduled mapping operations
- Frontend's MapAsync and Unmap operations become cheap state-only
  updates that do not interact with the D3D11 device context.

Key components:
- Buffer::TryMapNow(): Performs actual D3D11 mapping after GPU usage
  completes
- Queue::ScheduleBufferMapping(): Schedules buffer for re-mapping after
  queue submission
- Queue::CheckScheduledBufferMappings(): Processes and executes
  scheduled buffer mappings
- mAutoMapMode: Stores the computed map mode (Read/Write/Both) per
  buffer based on usage flags

Benefits:
This optimization eliminates lock contention between the frontend's
frequent MapAsync/Unmap calls and the backend's Queue::Submit
operations. By centralizing all D3D11 mapping operations within the
Queue's execution context, the frontend no longer needs to acquire the
D3D11 device context lock during map/unmap operations, significantly
reducing synchronization overhead.

Bug: 481158754
Change-Id: If8ebcad133c73db4b0b5b65d56b19cf63c22d362
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/286675
Commit-Queue: Quyen Le <lehoangquyen@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>