The caching layer's controller could not keep itself up to date

The component that manages the shared caching layer had been running the same version since last year, four releases behind, because the automatic update checks it relies on had never actually been able to run — the scheduled trigger they depend on did not exist.

Two further steps in its deployment had also never worked:

• the teardown step could never match what it was meant to remove, so it always failed
• the update step reported success on every run while doing nothing at all, because it looked for the deployment under a name that is never used

All three are fixed, and the controller now runs its current release on both clusters. Nothing about how services use the cache changed.