Test proposer lookahead initialization uses active validator set at fork

[linoscope]

Add a spec-test that checks whether proposer lookahead uses the correct active validator set at fork transition.
Had this test existed earlier, it would have detected the Prysm bug fixed by @potuz in OffchainLabs/prysm#15450.

Closes #4412 .

Validation note

To verify that the new test catches the Prysm issue, I temporarily replaced the reference implementation of initialize_proposer_lookahead with a buggy variant that mimics the original Prysm logic:

# Correct implementation (passes the new test)
def initialize_proposer_lookahead(
    state: electra.BeaconState,
) -> Vector[ValidatorIndex, (MIN_SEED_LOOKAHEAD + 1) * SLOTS_PER_EPOCH]:
    """
    Return proposer indices for the entire lookahead window starting at the current epoch.
    Used to initialise ``proposer_lookahead`` at genesis and after forks.
    """
    current_epoch = get_current_epoch(state)
    lookahead: list[ValidatorIndex] = []
    for i in range(MIN_SEED_LOOKAHEAD + 1):
        lookahead.extend(get_beacon_proposer_indices(state, Epoch(current_epoch + i)))
    return lookahead
    
 # Bug-replicating implementation (should fail the new test)
def initialize_proposer_lookahead(
    state: electra.BeaconState,
) -> Vector[ValidatorIndex, (MIN_SEED_LOOKAHEAD + 1) * SLOTS_PER_EPOCH]:
    """
    BUGGY: Re-uses the pre-fork active-validator set.
    """
    current_epoch = get_current_epoch(state)
    lookahead: list[ValidatorIndex] = []
    indices = get_active_validator_indices(state, Epoch(current_epoch))  # BUG: stale set
    for i in range(MIN_SEED_LOOKAHEAD + 1):
        seed = get_seed(state, Epoch(current_epoch + i), DOMAIN_BEACON_PROPOSER)
        lookahead.extend(
            compute_proposer_indices(state, Epoch(current_epoch + i), seed, indices)
        )
    return lookahead

Running the suite with the buggy version causes
test_proposer_lookahead_init_at_fork_only_contains_active_validators to fail, confirming that the new test detects the bug.

[jtraglia]

LGTM, nice and simple. Thank you!

[jtraglia]

Follow up. I can confirm that Prysm (dev branch) currently fails this test while all other clients pass it 😄

[linoscope]

Follow up. I can confirm that Prysm (dev branch) currently fails this test while all other clients pass it 😄

Cool to hear, thanks! That confirms that the test properly detects the bug & that other clients don't have the same bug 🙂