[ThoSap]

Fixes #97

CRStatus.setPhase was a setter with logic: when the phase changed, it set lastPhaseTransitionTime to now. Jackson also called it while it read a status from the cluster. The API server sends lastPhaseTransitionTime before phase, and a new CRStatus starts as PENDING. So each read of a resource that was not PENDING overwrote the stored time with now, and the next patchStatus wrote it back. On a test cluster, a ClusterConnection and 16 Grant resources that were READY since September all showed the time of the last operator restart.

What changed

1. setPhase is now updatePhase. Jackson does not use it as a setter, so it sets the phase field directly and keeps the stored time. The logic is the same. A Javadoc gives the reason. This is the same fix as in the Garage Operator (aboutbits/garage-operator@63da164).
2. All callers use updatePhase. 24 calls in the reconcilers and BaseReconciler, and 15 in the tests.
3. The Javadoc no longer says that a phase change sets the message to null. The code never did that.
4. New CRStatusTest. It reads a status JSON with lastPhaseTransitionTime before phase and expects the stored time. Other tests check that the same phase keeps a stored time, that a new phase replaces it, and that updatePhase can be chained with the other setters.

Consequences

• The generated CRDs do not change.
• After the upgrade, lastPhaseTransitionTime keeps its value until the phase really changes. Resources that the old version read show the time of that read, not of their last real phase change.

Checks that CI cannot do

• Without the fix, the deserialization test fails: the stored time 2026-01-01T00:00:00Z becomes the current time.
• The six generated CRDs are identical to the CRDs of the unfixed code.