Warm-up proposer preferences are gossiped before peers subscribe to the gloas topics #794

From the epoch before gloas, the validator client publishes proposer preferences for the first gloas epoch (#707). On a network with Lighthouse, those warm-up preferences reach no peer:

• ethlambda's beacon node joins the gloas topics an epoch before the fork and gossips each accepted preference at once;
• Lighthouse subscribes to them only 2 slots before the fork.

On #707's devnet (gloas at epoch 6), no Lighthouse node and no builder saw any of the warm-up preferences. None of our validators had a preference for the first gloas epoch, so builders' bids for those slots were ignored.

Nothing sends them again

• The validator client takes the 200 as delivered. feat(validator): publish proposer preferences (gloas) #707's once-an-epoch re-send does not help: the node answers an identical entry with 200 and does not republish it.
• The node logs Published proposer preferences to gossipsub at info even when the publish fails. The publish error is logged only at debug (swarm_adapter.rs).

Possible fixes

• Republish the pooled preferences for the fork epoch at the fork boundary, or once the gloas topics have peers.
• Have the node republish an entry it holds whose earlier publish failed, when the client re-sends it.
• Publish the warm-up epoch's preferences closer to the fork.

Separately, log a failed gossip publish at warn rather than debug.