add cirun-openstack-gpu-large to tinygrad

[h-vetinari]

This is another attempt at #1818, by using one of the openstack runners (it's similarly useful to have have an actual GPU to test the JIT code there, so this makes sense independently).

That is because

admin-requests/conda_forge_admin_requests/access_control.py

Lines 162 to 168 in fd804c9

	# this part is specific to github.com/Quansight/open-gpu-server
	if all(resource.startswith("cirun-openstack") for resource in resources):
	for key, value in DEFAULT_CIRUN_OPENSTACK_VALUES.items():
	for arg in value:
	register_ci_cmd.extend((f"--{key.replace('_', '-')}", arg))
	elif all(resource.startswith("cirun-") for resource in resources):
	pass

does not actually work for non-openstack runners. I had not understood this aspect at the time of da8b21e, apologies.

This is not easy to extend because the interface in smithy expects --cirun_users_from_json=... as an argument, which is not necessarily what we want for other resources (where a simple users: list should be created). We'll probably have to expand that interface (both in the yamls here, as well as in smithy; cirun itself should be generic enough AFAICT).

Concretely, after #1818 there was a PR to the feedstock (though not to .cirun, which I handled in conda-forge/.cirun#138), but it did not set up any actions.

This is yet another aspect of #1749