A GPU buffer that can act on itself — built via
BufferBuilder, never constructed
directly. Carries its own device/queue access internally, so writing to
it doesn’t need a &wgpu::Queue threaded in from the caller.
A buffer sized to hold many dynamically-offset elements — built via
DynamicBufferBuilder. Bundles
the per-element stride (for write_element) and
the true (unpadded) element size (for
BindGroupBuilder::dynamic_buffer)
alongside the buffer itself, so neither can drift out of sync with what
the buffer was actually built with.