feat: Add a way to use dynamic and static offsets for buffers in pipelines

WebGPU lets us bind parts of buffers to get at memory more efficiently and not make tons of extra buffers. Right now, our pipelines make us bind the whole buffer every time, unless you hack it by giving the same GPU buffer to a bunch of TgpuBuffers with different schemas (but they will always start from offset 0). The most obvious use-case for this is buffers with array schemas. We could add a way to make a "view" on a piece of a buffer that has an array schema when binding it. (There is no clear section in the spec about offsets so here is some reference)

Start with the WebGPU dynamicOffsetOrder reference linked in the issue, then inspect the pipeline buffer-binding flow and TgpuBuffers behavior described there. Compare how array schemas are currently bound and define what static and dynamic offsets, including buffer views, should support. Done means pipelines can bind the intended buffer sections without requiring separate buffers or offset-zero workarounds.