Memory cache never compacts expired keys with a stable set of cache keys

We're using Benthos to periodically monitor and process a set of IDs. To prevent needless reprocessing, we're caching IDs for some period of time. If the ID wasn't successfully processed downstream after the cache key TTL, we expect the ID to have been purged from the cache and be reprocessed.

Minimal configuration:

input:
  generate:
    interval: 1s
    mapping: root.key = "a"
pipeline:
  threads: -1
  processors:
    - cache:
        resource: mycache
        operator: add
        key: '${! json("key") }'
        value: "foo"
        ttl: 10s
    - mapping: root = if errored() { deleted() }
output:
  label: ""
  stdout:
    codec: lines
cache_resources:
  - label: mycache
    memory:
      compaction_interval: 1s

Expected output after 10s

{"key":"a"}
{"key":"a"}

Actual output after 10s

{"key":"a"}

The current work around is to always set a key on each pass through the pipeline to force a compaction for expired cache entries.

Start by running the minimal YAML configuration in the issue and observe the memory cache after the 10-second TTL with a stable key. Trace the memory cache compaction behavior and verify that expired keys are removed and the ID is emitted again without forcing a new key write.