test-driver: align fork-choice Hive step validation

[zclawz]

Summary

Fixes the zeam-specific failures from Hive lean-spec-tests-fork-choice run:
https://hive.leanroadmap.org/suite.html?suiteid=1784614222-c06cf50d0d1f10e59de4a259d1c33263.json&suitename=lean-spec-tests-fork-choice&client=zeam_devnet5

The run reported 9 zeam_devnet5 failures:

• test_block_future_horizon/test_block_at_clock_horizon_edge_imported
• test_block_future_horizon/test_block_beyond_future_horizon_rejected
• test_block_future_horizon/test_block_one_past_horizon_rejected
• test_early_block_arrival/test_block_ahead_of_store_clock_is_imported
• test_gossip_aggregated_empty_participants/test_gossip_aggregated_attestation_empty_participants_rejected
• test_gossip_aggregated_registry_and_signature/test_aggregated_attestation_participant_outside_registry_rejected
• test_gossip_aggregated_registry_and_signature/test_aggregated_attestation_proof_verification_failure_rejected
• test_gossip_attestation_validation/test_attestation_slot_near_uint64_max_rejected
• test_tick_acceptance_branches/test_interval_0_acceptance_with_proposal_recomputes_head

Fix

• Make the HTTP fork-choice test driver honor block-step tickToSlot=false, matching the local spectest runner. This keeps the driver clock stationary for early block arrival fixtures and rejects blocks beyond MAX_FUTURE_SLOT_TOLERANCE.
• Make tick steps honor hasProposal / has_proposal, so proposal-bound acceptance at slot boundaries can recompute head.
• Validate aggregated gossip participants before mutating fork-choice: reject empty aggregates, participant bits outside the validator registry, and fixture-declared signature/proof failures.
• Store accepted aggregated gossip payloads with .gossip attribution instead of .block_payload.
• Make fork-choice gossip attestation future-slot validation overflow-safe for near-u64.max slots, returning AttestationTooFarInFuture instead of letting request handling become HTTP 400.

Validation

• zig fmt --check pkgs/cli/src/test_driver.zig pkgs/node/src/forkchoice.zig
• zig build --fetch --summary none
• git diff --check

Full build/test not run in this container: PATH has Zig 0.15.2 while the repo requires Zig 0.16.0. I attempted to fetch Zig 0.16.0 side-by-side under /tmp, but the download was too slow for this Telegram turn.

[zclawz]

Done: PR #1055 is open for the fork-choice Hive failures.

PR: #1055
Commit: 4876576a8d3db4e820325e5a2b70dd2e009b6540

Validation completed locally:

• zig fmt --check pkgs/cli/src/test_driver.zig pkgs/node/src/forkchoice.zig
• zig build --fetch --summary none
• git diff --check

Full build/test note: this container has Zig 0.15.2 on PATH while zeam now requires Zig 0.16.0. I attempted to fetch Zig 0.16.0 side-by-side under /tmp, but the download was too slow for this Telegram turn, so I did not claim a full compile/test run.

[zclawz]

Testing update with Zig 0.16.0 completed.

Toolchain used:

• /tmp/zig-x86_64-linux-0.16.0/zig version → 0.16.0

Passed locally:

• /tmp/zig-x86_64-linux-0.16.0/zig fmt --check pkgs/cli/src/test_driver.zig pkgs/node/src/forkchoice.zig
• git diff --check
• /tmp/zig-x86_64-linux-0.16.0/zig build --fetch --summary none

Full test attempted:

• timeout 3600 /tmp/zig-x86_64-linux-0.16.0/zig build test --summary all

Result:

• Failed with exit code 1 because this container does not have rustup.
• Root error from the log:
  • error: failed to spawn and capture stdio from rustup: FileNotFound
  • failed command: rustup run nightly cargo -C rust -Z unstable-options build --profile multisig-release -p zeam-glue --no-default-features --features hashsig,multisig,jemalloc

Useful partial results before the transitive rustup failure:

• preset loading: 1/1 passed
• utils tests: 33/33 passed
• metrics tests: 7/7 passed
• params package: success
• xmss shadow-cost package: 6/6 passed

So the prior Zig-version blocker is fixed: PR #1055 was tested with Zig 0.16.0. The remaining local full-suite blocker is missing rustup, which prevents the Rust STF glue build and causes dependent Zig packages to fail transitively.

[ch4r10t33r]

LGTM.

Production change is a clean overflow guard: std.math.mul(data.slot, INTERVALS_PER_SLOT) catch -> AttestationTooFarInFuture, so a near-u64.max attestation slot is rejected gracefully instead of panicking/wrapping.

The rest is the Hive fork-choice HTTP test driver (not production code): honor tickToSlot=false (stationary clock + saturating future-horizon reject), honor hasProposal on tick steps, and validate aggregated gossip before mutating fork-choice (empty bits, out-of-registry participants, fixture-declared sig/proof failures), storing accepted payloads with .gossip attribution instead of .block_payload. Logic mirrors production's verifyAggregatedAttestation gates and the local spectest runner; indices computed once and freed. Test-harness only, cannot affect the running node.

CI fully green (build + test both platforms, gen-fixtures, provers, docker) — confirms the new driver code compiles and no spectest regression, which the author could not verify locally.