fix: persist cleaned epoch transition state before ack

[matthias-wright]

Addresses: #246 and #325

Changes

• Move finalized-block ack after epoch-boundary state persistence.
• Clear added_validators / removed_validators before storing next-epoch consensus state.
• Add a restart regression test that verifies persisted transition deltas are cleared and the active validator set is correct after reload.

Note: once this and #382 are on the same base, we should add a finalizer test that injects a commit failure at an epoch boundary and asserts the syncer ack is withheld (the Exact waiter resolves Err/canceled) and the node shuts down without durably advancing the epoch.

[sebastian-osec]

This fixes the main #246 persistence issue: transition deltas are cleared before the next-epoch consensus state is stored and committed. One follow-up concern: the finalized-block ack now happens after the oracle/orchestrator epoch-transition reports as well as after DB commit. Is that intended? If oracle/orchestrator reporting can block, this moves external backpressure onto the finalized-block ack path even after durable persistence has succeeded.

[matthias-wright]

Update(7077fd6):

• Send finalized block ack before the oracle/orchestrator reports.