fix: recover raft bootstrap from coordinator nodedown races

[cpluss]

Summary

• make Raft bootstrap convergence-driven instead of one-shot by adding retry scheduling when startup sync is incomplete
• trigger sync retries on write-node nodeup and keep startup completion idempotent so repeated sync passes do not reset readiness state
• harden coordinator group bootstrap by requiring visible replicas, surfacing bootstrap errors without crashing, and reusing a public RaftHealth.consensus_probe/0
• add regression tests for retry scheduling, retry cancellation on convergence, and startup-complete idempotency

Verification

• mix precommit

[cpluss]

Validation update on latest commit f16e103:

• mix precommit passes locally (191 tests, 0 failures, 3 skipped).
• CI checks are green:
  • build-and-push
  • test
  • test
• Local docker integration validation completed after explicit image build:
  • docker compose -f docker-compose.integration.yml -p raft-local build
  • docker compose -f docker-compose.integration.yml -p raft-local up -d
• Fresh bring-up converged:
  • /health/ready on node1 returned 200 {"status":"ok","mode":"write_node"}.
  • Ops.status() on node1/node2/node3 all reported local_ready: true, with all three nodes present in ready_nodes and raft_ready_nodes.
• Restart drill passed:
  • Restarted node1 and reconverged to all three nodes ready.
• RYW check passed:
  • Write on node1 (create session + append seq=1).
  • Immediate reads on node2 and node3 saw the appended event (seq=1, payload text "ryw-check").

Branch is clean and fully pushed.