[Bug]: HTTP cluster get_prefix exposes expired keys before cleanup sweep

System Info

Platform-independent HTTP cluster-storage behavior on current main.

Reproduction

HttpClusterStorageServer.get() checks expire_time before returning a value, but get_prefix() returns every matching entry without checking expiration. The background cleanup task runs periodically, so during the interval after a TTL expires and before the next sweep, get_prefix() can return stale data that get() already considers expired.

Expected behavior

Point reads and prefix reads should use the same expiration semantics. Expired entries must not be returned by get_prefix().

Actual behavior

Expired entries remain visible through prefix reads until the background cleanup loop removes them.

Proposed fix

Filter prefix results using the same expire_time rule as get(), with a deterministic CPU-only regression test that does not depend on sleeping for the cleanup interval.

Start at HttpClusterStorageServer.get_prefix() and compare its expiration handling with get(), focusing on the existing expire_time rule. Add a deterministic CPU-only regression test for prefix reads with an expired entry, without waiting for the cleanup interval. Done means get_prefix() no longer returns entries that get() considers expired.