The introspection cache used the raw token as key in a shared cache.
Two instances with different providers or clients shared results. Keys
now use the same instance prefix as the verified-token cache.
The verified-token cache is one process-wide cache keyed by the raw
token. A verdict stored by one middleware instance satisfied any other
instance, so a token valid for one client or audience skipped the
issuer, audience and signature checks on another route. Each instance
now prefixes keys with its provider, client ID and audience. Dynamic
client registration updates the prefix.
lukaszraczylo