Fix issues with calling resolve_queries multiple times per frame

[Wumpf]

• Fixes wgpu alignment validation errror on GpuProfiler::resolve_queries #79

Turns out there were two issues!

The more glaring one was that wgpu-profiler ignored QUERY_RESOLVE_BUFFER_ALIGNMENT.
This however only happens when calling resolve_queries multiple times a frame: query pools are not shared across frame (sharing across frames would make it impossible to get results of a single frame in unison and would lead to awkward stitching), so the only way to have num_resolved_queries be !=0 is when there's multiple resolves in a single frame.
And sure enough a simple tests catches this!

Looking closer still though it doesn't make sense we ever use any offset at all since the result is right away copied into the "read buffer" which doesn't have such strict alignment criteria (merely wgpu::COPY_BUFFER_ALIGNMENT which is 4). This in turn reveals that HERE we always used an offset of 0 meaning that previous results were overriden 😬
Unfortunately, I don't know how to reliably unit test this part of the issue :(

[Wumpf]

Confirmed working here #79 (comment)