cluster_yaml is not cleaned up

[cg505]

During review for #4980, it was noted that CloudVmRayBackend.remove_cluster_config has a bug that prevents it from actually deleting the config:

handle.cluster_yaml = None
global_user_state.update_cluster_handle(handle.cluster_name, handle)
common_utils.remove_file_if_exists(handle.cluster_yaml) # but cluster_yaml was already set to None...

The fix is obvious, but we should be careful since it could introduce a leak if other code was unintentionally relying on this.

While looking at this, I noticed in post_teardown_cleanup that we call remove_cluster_config in the terminate case

skypilot/sky/backends/cloud_vm_ray_backend.py

Line 4294 in a7f9295

self.remove_cluster_config(handle)

and then later there is an if statement that depends on handle.cluster_yaml...

skypilot/sky/backends/cloud_vm_ray_backend.py

Lines 4360 to 4361 in a7f9295

	if handle.cluster_yaml is not None:
	_detect_abnormal_non_terminated_nodes(handle)

So we will never hit this.

We need to look carefully at this function and actually specify the expected invariant for when handle.cluster_yaml will/won't be set and when the file will/won't exist.