…releases

cozystack-api writes healthCheckExprs and the resolved wait strategy
from the ApplicationDefinition only when it builds a HelmRelease, and
the definition reconciler that keeps existing releases in step copied
neither. A release created before its definition exempted an object
kept waiting on that object at every upgrade. For vm-disk that is
every disk created before the DataVolume expression was added: a disk
whose import is still running after a platform upgrade fails the
upgrade at the timeout and, under RetryOnFailure, retries forever
while holding a helm-controller worker.

The reconciler now derives both fields the way the API does, in both
directions, so an existing release ends up with what a fresh write
would produce. On the shipped definitions this only relaxes waits: the
vm-disk DataVolume and the postgres KEDA ScaledObject. A release that
another chart renders carries Flux's origin label and keeps the gate
its chart sets.

A failed update used to be logged and dropped, and nothing else
re-enqueues the definition for hours, so that release stayed out of
sync. The errors are now returned and the definition is retried.

Fixes #4409

Assisted-by: LLM
Signed-off-by: Myasnikov Daniil <[email protected]>
(cherry picked from commit 8d8179f)