fix: lazy featuregate init in callers

[estroz]

Fixes #702

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[jgehrcke]

Instantiating the feature gate singleton only if there's actually a consumer: appears to be a tidy approach. @estroz you tested that this actually fixes #702 for you, right?

We have identified an invariant here that want to maintain. Hence, we may want to add a pragmatic test to make sure that we don't regress from here. We can do that in another PR.

[klueska]

There was a desire in the initial design to not have to call a function, but rather just access a global variable from the featuregate package (similar to as is done in k8s proper). I'm not 100% opposed to this approach, but I would want to then remove the global FeatureGates var and then call the function just FeatureGates() instead of GetFeatureGates().

[estroz]

@jgehrcke yes I have confirmed this fixes the panic for type imports.

@klueska renamed the function and accepted the suggestion, thanks.

[klueska]

I think we can still get away with having FeatureGates be a global variable instead of a function. We just need to implement its methods on its pointer type and init it with the once iff the incoming pointer is nil and equals the address of the global FeatureGates variable.