Fix: Register all BFT sub-protocols during IBFT2→QBFT consensus migration

[saeeddawod]

Summary

This PR fixes a critical bug that prevents successful IBFT2 → QBFT consensus migration in Besu. During migration, nodes would stall and fail to produce blocks because they could only speak one BFT wire protocol instead of both.

Problem

When running a consensus migration from IBFT2 to QBFT using a transitions configuration in genesis, the network would randomly stall. The migration worked sometimes by luck, but failed other times.

Root Cause

In ConsensusScheduleBesuControllerBuilder.createSubProtocolConfiguration(), the original code was:

return besuControllerBuilderSchedule
    .get(besuControllerBuilderSchedule.keySet().stream().skip(1).findFirst().orElseThrow())
    .createSubProtocolConfiguration(ethProtocolManager, maybeSnapProtocolManager);

Issues with this approach:

1. Non-deterministic ordering: besuControllerBuilderSchedule is a HashMap, so keySet().stream().skip(1) produces unpredictable results depending on hash ordering
2. Only one protocol registered: Only the sub-protocol configuration from ONE consensus mechanism was registered
3. Random failures: If HashMap ordering picked QBFT protocol while consensus was still IBFT2, nodes couldn't exchange consensus messages → network stalls. Sometimes by luck it would pick the matching protocol and work.

Solution

The fix registers all BFT sub-protocols from all scheduled consensus mechanisms:

• IBF/1 (IBFT2) - for pre-migration communication
• istanbul/100 (QBFT) - for post-migration communication

This allows nodes to communicate using both protocols during the transition period. The fix:

1. Iterates through all consensus builders (sorted by block number for determinism)
2. Merges their sub-protocol configurations
3. Avoids duplicates (e.g., EthProtocol which is common to both)

Testing

Migration Test Results

Phase	Block	Consensus	Status
Pre-fork	#0-49	IBFT2	✅ IbftBesuControllerBuilder
Fork	#50	QBFT	✅ QbftBesuControllerBuilder
Post-fork	#50-100+	QBFT	✅ Continuous block production

Log Evidence

# Pre-fork (IBFT2)
IbftBesuControllerBuilder | Produced #49 / 0 tx / ...

# Fork block (QBFT takes over)  
QbftBesuControllerBuilder | Imported empty block #50 / 0 tx / ...

# Post-fork (QBFT continues)
QbftBesuControllerBuilder | Produced empty block #73 / 0 tx / ...

Checklist

• Checked out contribution guidelines
• Considered documentation - migration docs may need update
• Considered changelog - This is a bug fix for consensus migration
• No database changes

Local Tests

• Manual migration test (IBFT2 → QBFT) - PASSED
• spotless: ./gradlew spotlessApply
• unit tests: ./gradlew build
• acceptance tests: ./gradlew acceptanceTest

[macfarla]

@saeeddawod can you add a changelog entry since this is a bugfix?

[saeeddawod]

@saeeddawod can you add a changelog entry since this is a bugfix?

@macfarla done !

[macfarla]

LGTM

[macfarla]

@saeeddawod spotless check is failing, and I can't push to your branch to fix it - can you run spotlessApply gradle target. Assigning back to you.

[saeeddawod]

Thanks @jframe @macfarla Addressed the review:

• Fixed imports (HashSet, Set instead of fully qualified names)
• Added createsSubProtocolConfigurationWithAllSubProtocols() test
• Ran spotlessApply