fix(validator): remove duplicate time-sensitive validation of block proposals

[spalladino]

Removes the duplicate block-proposal validation that ran again downstream of p2p ingress, whose repeated wall-clock check could turn an honest, on-time proposal into a false slashable offense.

Context

On mainnet a validator node classified a canonical block proposal as a slashable invalid_proposal offense. Block proposals are validated at p2p ingress, where the receive-window check decides whether they arrived in time; ProposalHandler.handleBlockProposal then re-ran that same validation "out of caution" before re-execution. When node-local processing stalled (~50s in the incident), the repeated wall-clock check failed an on-time proposal, and invalid_proposal is in SLASHABLE_BLOCK_PROPOSAL_VALIDATION_RESULT — so local latency was converted into slashing evidence against an honest proposer.

Approach

Delete the downstream blockProposalValidator.validate(proposal) call and, with it, the now-unused BlockProposalValidator dependency of ProposalHandler (and its construction in ValidatorClient.new and createProposalHandler).

This is safe because every path into handleBlockProposal is post-ingress — gossiped block proposals and checkpoint-embedded blocks both pass ingress validation first, and there is no replay path — so the second pass added no coverage. The checks that remain downstream are deterministic properties of the signed payload rather than of the clock: the signature via getSender(), duplicate tx hashes, checkpoint/index consistency, in-hash agreement, embedded-tx integrity via tx collection, and full re-execution. The checkpoint-proposal path already operates without such a re-check. invalid_proposal stays in the slashable list, since it is still produced by the structural computeCheckpointNumber checks. The p2p package is untouched: ingress keeps its own validator.

Note that dropping the whole call also stops re-deriving the expected proposer downstream; that identity check likewise belongs to ingress, and the two validator.test.ts cases that pinned the old repeated-gate behavior were rewritten to assert the new contract (validates, no WANT_TO_SLASH_EVENT, slot not marked invalid).

This is the simpler alternative to #25201, which instead splits the validator into validate() / validateStableFields() and adds a non-slashable reason; the two are open in parallel for comparison.

Fixes A-1703

[AztecBot]

❌ Failed to cherry-pick to next due to conflicts. (🤖) View backport run.