Cope when syncing a helm-provided CRD and resources using the CRD

[brantb]

#1825 describes an issue where a commit cannot be synced if it contains both a CRD and a resource that uses that CRD. A fix was implemented in #1876 to look in the git repo for the manifest, however, this does not work if the CRD is installed via helm.

To reproduce:

• Have a git repository containing a HelmRelease for rbac-manager and an RBACDefinition manifest
• Create a new Kubernetes cluster
• Install flux (v1.12.0) and flux-helm-operator

Expected result:

Flux applies the HelmRelease, which creates the RBACDefinition CRD, which allows Flux to apply the RBACDefinition manifest. In flux <= 1.10.1 this would take multiple sync passes, but the cluster would eventually reach the desired state.

Actual result:

After cloning the git repo, flux writes the following message to the log, and does not sync anything:

s=2019-04-15T19:43:02.307951117Z caller=images.go:24 component=sync-loop error="getting unlocked automated resources: error looking up API resources for RBACDefinition.rbacmanager.reactiveops.io/v1beta1: not found"

Since the CRD is part of the rbac-manager helm chart the code path added in #1876 doesn't seem to find it.

[2opremio]

This one's tricky and calls for implementing the approach suggested by @squaremo in #1825 (comment)

[2opremio]

BTW, the workaround here is to first add the HelmRelease, wait for it to sync and then add the CRD instances. Sorry for the inconvenience.