fix: retry conclude_node_migration until contract reflects completion, not until tx is applied

[anodar]

Work towards: #2898

retry_conclude_onboarding previously treated tx_sender::send returning Ok as success, but that only signals the tx was applied — NEAR rolls back state when a contract method returns Err, so a contract-side rejection (e.g. InvalidTeeRemoteAttestation racing the new node's attestation submission) silently left the migration record in place and the onboarding node waiting forever on a participant list change that would never happen.
Now we verify the migration actually completed by waiting for active_migration to flip false on the local MigrationInfo watch — set when the contract removes our migration record on success — and retry the conclude tx if it doesn't.

Verified by running all tests 100 times in #3024, specifically this run: https://github.com/near/mpc/actions/runs/25051806158/job/73381273673

[claude]

Code Review

No critical issues found. The change correctly adds a post-tx verification step to retry_conclude_onboarding by watching for active_migration to clear, addressing the case where tx_sender::send returns Ok even when the contract method rolled back.

The logic is sound:

• watch::Receiver::clone() inside each retry attempt correctly gets a fresh view of the current state
• The borrow_and_update() check before changed().await handles the already-cleared case
• Timeout + retry is a reasonable approach for observing eventual consistency

Minor note (non-blocking): wait_for_active_migration_to_clear is a nicely isolated function that would benefit from unit tests (timeout path, already-cleared path, channel-closed path), but the logic is straightforward enough that this doesn't block merge.

✅ Approved

[kevindeforth]

Thanks a lot! I remember skipping this part when implementing the migration logic.

edit: would be curious to know why the transactions fail though. I don't think this was an issue in the pytests.

[netrome]

Nice!