fix(controllers): retry storage updates on conflict, log other errors

[indyjonesnl]

Problem

Four controllers used `let _ = self.storage.update(...).await` to write status updates, silently swallowing every storage error. On a benign resource-version `Conflict` the update vanished — the controller's view of reality diverged from etcd. On a hard error nothing was logged, so failures were invisible.

Fix

Each controller now matches `Err` explicitly:

• apiservice: `update_condition` writes the `Available` condition. On Conflict, re-read and re-apply once. Other errors log via `tracing::error!` with the apiservice name.
• statefulset: Three `reconcile()` sites set `deletionTimestamp` on pods (terminal-pod cleanup, scale-down, rolling update). Routed through a new private `update_pod_with_retry()` helper: re-read + reapply once on Conflict, no-op on NotFound, log other errors.
• scheduler: Six `let _ = ...` sites covering priority resolution, nominatedNodeName fallback after failed binding, and PodScheduled=Unschedulable condition writes. Same helper pattern.
• deployment: Three improvements bundled because they all hit `reconcile_deployment`:
  1. Drop `tokio::task::block_in_place(|| Handle::current().block_on(...))` from the available-replica fallback — the surrounding fn is already `async`. Replace with a sequential `.await` accumulator.
  2. Add exponential backoff with jitter (10ms → 500ms) to the deployment-revision CAS retry. Without jitter, contended deployments thrashed storage.
  3. Convert four silenced `let _ = self.storage.update(...)` sites (orphan-RS adoption, RS/deployment revision annotation refresh, RS replica annotations) to log non-Conflict errors via `tracing::error!`.

Verification

`cargo build --workspace --locked` clean. Full test suite is green when stacked on #32; this PR adds no new test failures relative to the baseline.

Dependency

Depends on #32 for the test suite to show green.