QRM returns success when checkpoint write fails

[AutuSnow]

What happened?

CPU and memory QRM policies only log checkpoint write errors. Allocate and RemovePod can still return success.

After a restart, a successful allocation may disappear, or removed pod state may come back.

How to reproduce

1. Make cpu_plugin_state or memory_plugin_state unwritable.
2. Call Allocate or RemovePod.
3. The checkpoint write fails, but the RPC still succeeds.
4. Restart the policy and it restores the stale checkpoint.

Expected behavior

Return an error when the final checkpoint update fails.

[AutuSnow]

/assign