Initial refactor to how bind groups can be used with non-persistent UBs

[mvaligursky]

• related to the suggested design here: WebGPU: BindGroups aren't being reused #6315 (comment)
• this is a first try at doing this, and applied to WebgpuClearRenderer only, but it works.
• basically each dynamic buffer from which we bump allocate non-persistent uniform buffers has a cache of bind groups based on uniform buffer size (as this size is part of the bind group). And this means, all buffers using the same size can share a single bind group, and avoid their allocation each time the bump allocator switches to a different buffer
• also did a small refactor to the way uniforms are set to the UB, allowing users to avoid using the scope in the middle and set values directly to the buffer.

The next PR will apply this to the mesh UB, which is a main reason for the allocations.

[slimbuck]

adding these to the code is so incredibly useful <3