sylva-units lets a HelmRelease reconcile with new chart code but old values

I observed the following in https://gitlab.com/sylva-projects/sylva-core/-/jobs/8397904618:

HelmRelease/sylva-system/cluster-bmh              InProgress               Running 'upgrade' action with timeout of 5m0s
╰┄╴┬┄┄[Conditions]
   ├┄╴Reconciling                                 True       Progressing   Running 'upgrade' action with timeout of 5m0s
   ├┄╴Ready                                       False      UpgradeFailed 
Helm upgrade failed for release sylva-system/cluster-bmh with chart sylva-capi-cluster@0.0.0+3d0e10994448: values don't meet the specifications of the schema(s) in the following chart(s):

   ┆  sylva-capi-cluster:
   ┆  - (root): Additional property use_custom_rancher_dns_resolver is not allowed
   ╰┄╴Released                                                                     False UpgradeFailed 
    Helm upgrade failed for release sylva-system/cluster-bmh with chart sylva-capi-cluster@0.0.0+3d0e10994448: 
values don't meet the specifications of the schema(s) in the following chart(s):
      sylva-capi-cluster:
      - (root): Additional property use_custom_rancher_dns_resolver is not allowed

This is an "upgrade from Sylva 1.1.1" run.

The above show that we have:

• cluster-bmh HelmRelease using sylva-capi-cluster 0.4.2 (commit 3d0e10994448 is 0.4.2) which is the version in main
• but in values we have use_custom_rancher_dns_resolver which is present only in Sylva 1.1.1

So we have "old values" used with "recent code", which is an anomaly.

This seems pretty rare so I suspect some race in when/how the HelmRelease and GitRepository are updated.

/cc @feleouet