Resource names exceed the 63-char limit with long release/override names

Several templates use a fullname helper's output as a prefix and append suffixes. Truncation happens inside the helper, before the suffix is appended, so final resource names can exceed the 63-character Kubernetes name limit.

Reproduction:

helm template $(printf 'r%.0s' {1..30}) . --set nameOverride=longvaluehere \
  | grep -nE 'name:' \
  | awk '{ for(i=1;i<=NF;i++) if(length($i)>63) print length($i), $0 }'

This prints ~20 names over 63 chars, in three groups:

1. KMM subchart (kmm.fullname) e.g. -kmm-event-recorder-clusterrolebinding. Longest appended suffix: -event-recorder-clusterrolebinding (34 chars).

2. Top-level gpu-operator chart e.g. -event-recorder-clusterrolebinding, -metrics-exporter-rbac-proxy, -leader-election-rolebinding. Same root cause in the top-level fullname helper.

3. node-feature-discovery dependency -node-feature-discovery-{master,worker}-conf. Originates in the NFD subchart; likely not fixable in this repo.

You can see an example here.

Solutions

Possible solution: reserve room for the longest suffix in the fullname helpers, shorten the resource-name suffixes, or append a short hash to preserve uniqueness. Happy to implement a fix if needed.

Run the provided helm template reproduction and inspect helm-charts-k8s/templates/event-recorder-clusterrolebinding-rbac.yaml, then trace the top-level and KMM fullname helpers that produce the affected names. Confirm which rendered resources exceed Kubernetes's 63-character limit, implement a consistent fix for the top-level and KMM charts, and rerun the reproduction to verify those names are within the limit; the NFD dependency may remain out of scope.