test(validator-client): deflake integration test with frozen clock

[spalladino]

Motivation

validator.integration.test.ts (suite ValidatorClient Integration) has been flaking on both next and merge-train/spartan-v5 — failed ~7 times, e.g. rejects block that would exceed checkpoint mana limit. It is a true flake: a docs-only PR ran the same command 3×, failing once and passing twice.

Root cause: the block re-execution deadline is an absolute timestamp — the slot attestation deadline (target_slot_start + S − 2E ≈ 24s for the test's constants) — compared against the date provider's clock in the public processor (now() > deadline → "Stopping tx processing due to timeout"). The suite used TestDateProvider, whose now() advances with real wall-clock from the beforeEach anchor. So the ~24s slot-1 re-execution budget started ticking before the two heavy createValidatorContext setups. On a loaded CI machine, fixture setup (~22s observed) consumed the budget before block re-execution ran; a block that should re-execute successfully instead hit the deadline, processed 0 txs, and was rejected as an empty non-first block ("Cannot add empty block that is not the first block in the checkpoint") — the same rejection the mana-limit test expects only for the overflowing block.

Approach

• Switch the suite from TestDateProvider (drifts with real time) to the frozen ManualDateProvider, whose clock only moves on explicit setTime/advanceTime. Re-execution can no longer race wall-clock, so slow fixture setup never eats the per-slot budget.
• The two tests that rely on a retryUntil timing out (refuses to attest if not all block proposals were processed, refuses to attest with archive mismatch) would otherwise poll the full ~24s window in real time under a frozen clock. They now advance the clock past slot 1's attestation deadline before attestToCheckpointProposal, so the (correct) timeout fires immediately.

No assertion was relaxed; the behavioral checks are unchanged.

Changes

• validator-client (tests): use ManualDateProvider in validator.integration.test.ts; advance the frozen clock past the attestation deadline in the two timeout-dependent tests.

Verification

• Red repro: forcing the clock past the deadline between block 1 and block 2 reproduces the exact CI failure chain (timeout → "Cannot add empty block" → validateBlockProposal returns false).
• Green: all 8 tests pass; the two timeout-dependent tests dropped from full-window polling to ~0.8s.
• Root-cause robustness: injecting a 30s real sleep into beforeEach (which would blow the old 24s budget) still passes with the frozen clock.
• yarn build, yarn format, yarn lint clean.

Fixes A-1265