AccessTokenStore

Caches the platform access token of one channel credential.

Channel token providers consult this store before fetching a token from the platform and write each fetched token back. The provider owns the policy — refresh timing, response parsing, and error-code invalidation — and treats the store as an opaque cache slot: one store instance serves one credential, and the store is not keyed, so an instance shared across providers built from different credentials would silently mix their tokens.

The default InMemoryAccessTokenStore keeps the token in the JVM heap, so it is not shared across processes. Deployments running multiple channel instances should provide an implementation backed by shared storage (for example Redis keyed by channel id) and inject it through the channel factories: one instance's refresh or invalidation then serves the whole deployment.

Implementations must be thread-safe: providers invoke these methods from concurrent reactive pipelines. Concurrent refreshes are expected (there is no cross-instance lock); providers store fetched tokens with #putIfNewer so racing refreshes cannot regress the cache to an earlier refresh deadline, while #put replaces unconditionally.