Add inclusion list production to the validator client

[conache]

Issue Addressed

Add validator client support for producing inclusion lists for validators in the inclusion list committee (per spec).

Proposed Changes

• Add a new InclusionListService to the VC, which only starts when the Heze fork is scheduled and idles for whole epochs until Heze activates
• Each slot, 1s before the inclusion list deadline, the service checks the slot's IL duties and fetches the IL transactions from the BN, using GET validator/inclusion_list?slot={slot}
• For each duty, it builds a SignedInclusionList
• Each signed IL is published individually using POST validator/inclusion_list (beacon-APIs#490), SSZ first, using JSON as fallback
• This PR also includes tests for the service against MockBeaconNode and adds IL mock helpers to validator_test_rig

Additional Info

• The Beacon API endpoints are added in the following PRs:

  • Add inclusion list production endpoint to the Beacon API #10008
  • Add inclusion list publishing endpoint to the Beacon API #10174

• Heze types don't exist in remote-signing-api yet, so Web3Signer path is defined ahead of the upstream spec. This only follows the existing message-type setup and may need adjusting later, once the Heze support lands

• ILs are produced 1s before the IL deadline instead of at slot start. I chose this approach after trying multiple solutions. At slot start, the EL's mempool still holds the current payload's transactions, so ~92% of each IL was redundant (confirmed on a local devnet). Producing on payload reveal didn't help with this: the EL only drops those transactions after the next forkchoiceUpdated call. For this case, making the BN wait for it, although it could be a solution, adds complexity to the forkchoice update path. With this approach, tested on a local devnet, I've got no redundant txs included in the inclusion lists. Additionally, all the inclusion lists broadcast were timely. On FOCIL devnet-1 (based on 24h data), 2.7% of Lighthouse IL transactions were redundant, vs ~80% for clients producing at slot start, while 97.7% of (list, node) arrivals were timely, the late ones concentrated at a single slow node (lodestar-nethermind-1).