Add a TTL to ChainedFastBackend to prevent long running requests from using data that is too stale

This was initially added by beejeebus in #2248767-53: Use fast, local cache back-end (APCu, if available) for low-write caches (bootstrap, discovery, and config) → and approved by catch in #2248767-62: Use fast, local cache back-end (APCu, if available) for low-write caches (bootstrap, discovery, and config) → , but I'm splitting it into a separate issue, because an arbitrary TTL can introduce random test failures based on how fast testbot happens to be (see #2248767-112: Use fast, local cache back-end (APCu, if available) for low-write caches (bootstrap, discovery, and config) → ).

Do we want a TTL, or would we prefer long running processes to call reset() at the points that they control? AFAIK, HEAD doesn't have a TTL for any other static cache, but that shouldn't rule out us adding it here if we think it makes sense.