Add range_query tests to NVML test suite

[charlesbluca]

Adds a test similar to test_get_worker_monitor_info() to test_nvml.py for use in Dask-CUDA's gpuCI tests; the idea here is that we can hopefully catch issues like rapidsai/dask-cuda#634 preemptively.

• Closes #xxxx
• Tests added / passed
• Passes black distributed / flake8 distributed / isort distributed

[pentschev]

We probably need #4866 for this, but Distributed CI likely won't catch this because it doesn't have GPU support yet.

[pentschev]

As discussed with @charlesbluca earlier, it's probably best to wait until #4873 is merged and then add the following check to this test:

if nvml.device_get_count() < 1:
    pytest.skip("No GPUs available")

[pentschev]

@charlesbluca #4873 is in now, you can now test with that.

[charlesbluca]

Looks like that fixes things - thanks @pentschev! I'll merge upstream and add the check.

[jrbourbeau]

Thanks @charlesbluca! I noticed all six tests in this module require nvml.device_get_count() >= 1. In order to minimize code duplication could we just skip this entire module if nvml.device_get_count() == 0?

[charlesbluca]

I believe so - I discussed this briefly with @pentschev but wasn't 100% sure if we needed to do a check per-test to ensure that the tests have proper NVML initialization.

[jrbourbeau]

Fair point. FWIW my impression was that device_get_count handled initialization properly:

distributed/distributed/diagnostics/nvml.py

Lines 26 to 31 in 2bdec05

	def device_get_count():
	init_once()
	if nvmlLibraryNotFound or not nvmlInitialized:
	return 0
	else:
	return pynvml.nvmlDeviceGetCount()

But I'll let @pentschev confirm whether or not this is the case

[pentschev]

I just didn't want to risk introducing another issue, but if it works with that change I'm fine with doing that.

[jrbourbeau]

Thanks @charlesbluca, this is in