`wgpuInstanceWaitAny` is unimplemented

[birkskyum]

wgpuInstanceWaitAny panics at runtime (src/unimplemented.rs:230). The broader WGPUFuture machinery is also stubbed out — all async APIs return a null future (id 0), and instance capabilities report timedWaitAnyEnable: false.

wgpuInstanceProcessEvents works fine and callbacks do fire during poll, so the workaround is a poll loop with a flag:

// standard (works on Dawn)
callbackInfo.mode = WGPUCallbackMode_WaitAnyOnly;
wgpuBufferMapAsync(buffer, ...callbackInfo);
wgpuInstanceWaitAny(instance, 1, &futureInfo, UINT64_MAX);

// wgpu-native workaround
callbackInfo.mode = WGPUCallbackMode_AllowProcessEvents;
wgpuBufferMapAsync(buffer, ...callbackInfo);
while (!done) {
    wgpuDevicePoll(device, true, NULL);
}

This works but forces anyone targeting both Dawn and wgpu-native into #ifdef branches for every async operation (buffer mapping, submitted work done, etc). We hit this in MapLibre Native's WebGPU backend.

I realize the hard part is that wgpu-core doesn't have a future-multiplexing primitive — it has per-device device_poll and poll_all_devices but nothing like "wait until any of N futures completes." So this probably needs either a future registry layer in wgpu-native on top of the existing polling, or upstream support in wgpu-core.

Related: #510 (user hitting the panic), webgpu-headers #199 (the spec-level design for WGPUFuture, now resolved).