DAL/GS: Recompute shards proofs from shards if not in memory cache?

Currently there is a bounded memory cache for shard proofs. It is updated when the shards of a slot are computed and used when the shards are injected to GS (done when the slot header is waiting for attestation on L1).

Should we recompute the proofs from the shards when needed if they are not in the memory cache anymore?