Compute attestation aggregation in parallel with block production flow

[tbenr]

Computing attestation aggregation currently takes at least 200ms on holesky node.
We could implement a strategy to trigger the aggregation in parallel with the block production flow.

there is a spike PR #9122 demonstrating it could save us time during block production.
In that PR description I reported some thoughts.

Slow Block Production *** Slot: 3635966 start 1ms, preparation_on_tick +0ms, preparation_apply_deferred_attestations +0ms, preparation_process_head +337ms, retrieve_state +3ms, beacon_block_prepared +362ms, local_get_payload +16ms, beacon_block_created +1ms, state_transition +230ms, state_hashing +18ms, complete +0ms total: 967ms

The main benefits will be to parallelize head processing (preparation_process_head step) with attestations aggregation (which happens in beacon_block_prepared), by optimistically passing the state from our current head (processed into the next slot), making sure to rerun the aggregation again if, during the actual block preparation we selected a "compatible" state.

The "compatibility" boils down to shuffling\committeeSizes compatibility and isValid compatibility (no differences in AttestationDataValidator checks).

We could wait Pectra activation to see aggregation timings and evaluate if it is worthy going down to this path.