Remove use of `XPackLicenseState::isAllowedByLicense` for ML models

[rjernst]

XPackLicenseState is meant to be an opaque view of the current license level. The LicensedFeature class should be used to determine whether a feature is used, so that tracking is always. However, there are a few special methods on XPackLicenseState that currently avoid feature tracking. This issue is about one of those: isMachineLearningAllowedForOperationMode.

We should convert the one user of this method to using a proper LicensedFeature.

[elasticmachine]

Pinging @elastic/ml-core (Team:ML)

[benwtrent]

@rjernst, the only user of this method is when we check the license level of a remote cluster. CCR (isCcrAllowedForOperationMode) does the same exact thing, verifying that the remote cluster license supports the given feature.

Is there work in this area already in getting CCR to stop using isCcrAllowedForOperationMode?

I am not sure which feature we would be tracking regardless as the API calls + persistent tasks cover the actual feature usage, we are simply checking the license of the remote cluster.

[rjernst]

Thanks Ben, I realized I was thinking of the wrong method, I've corrected the title of this issue. I will fix the remote checkers (which I don't think need to actually be tracked). The one I meant for this issue to cover is the use in TransportInternalInferModelAction and TransportPutTrainedModelAliasAction of passing in a license level (which is not tracked) instead of having checking a LicensedFeature object.

[benwtrent]

@rjernst awesome, I will look into isAllowedByLicense usage by ML