webgpu: Fix handling of WGPU_WHOLE_SIZE and large u64s (#18142)

Most cases that need to combine two 32-bit words into one u64 get two
i32s, because that's what the WASM->JS calling convention gives us. But
one place, in the handling of WGPUBindGroupEntry, was explicitly loading
two u32s from memory, which would give us 0xFFFFFFFF instead of -1.
Fixing this fixes the handling of WGPU_WHOLE_SIZE. Since that's fixed,
also cherry-picked a change into webgpu_cpp.h that makes that the
default for wgpu::BindGroupEntry, as it should be.

Additionally, the logic which converted two i32s into a u64 was written
for two u32s, so it was broken for any number over 0x00000000_FFFFFFFF.
This is fixed by converting the i32s into u32s (using >>> 0) first.

Finally, the names are improved so that errors would have been more
apparent.