I need a deep consensus-level code review of the XDCIndia/go-ethereum GP5 fork compared against the official XinFinOrg/XDPoSChain devnet.

## EXACT ISSUE

The node gets stuck at block 4435 (V2 epoch switch block, round 1800) with "validators not legit" error.

**What happens:**
1. Local node computes masterNodes = 100 entries (after applying 8 penalties from HookPenalty)
2. Official devnet block header has validatorsAddress = 108 entries
3. CompareSignersLists fails because lengths differ (100 vs 108)

**Root cause hypothesis:**
- During bulk sync, XdcBulkSyncMode skips transaction execution for non-checkpoint blocks (including gap blocks like 3150)
- UpdateMasternodesFromHeader at gap block 3150 reads candidates from smart contract state via state.GetCandidates(statedb)
- But statedb is stale (parent state, not updated state) because transactions were skipped
- Result: snapshot stores 108 original genesis candidates instead of 100 (after penalties from previous epoch 2700-3599)
- At epoch switch 4435, HookPenalty computes 8 penalties for epoch 3600-4434
- calcMasternodes tries removeItemFromArray(candidates, penalties) but some penalty addresses are NOT in the stale 108-candidate list
- Result: masterNodes ends up with 100 entries (partial removal), but official devnet header has 108

**Secondary issue at V2 switch block 2700:**
- When syncing from scratch, initial() requires checkpoint header at 2700 but sync starts from 2701
- lastGapHeader at 2250 exists in DB but checkpointHeader at 2700 missing
- Returns error instead of deferring initialization

**Current workarounds applied (UNSAFE for mainnet):**
1. verifyHeader.go lines 167-186: Bypass validator/penalty check for chainId 5551
2. engine.go lines 411-420: Defer initialization when checkpoint header missing but currentHead > switchBlock

## FILES TO REVIEW

1. core/state_processor.go - bulk sync skip logic (lines 133-154)
2. core/xdc_sync_opts.go - XdcShouldSkipTxExecution
3. consensus/XDPoS/engines/engine_v2/verifyHeader.go - validator verification
4. consensus/XDPoS/engines/engine_v2/engine.go - calcMasternodes, HookPenalty, initial, UpdateMasternodesFromHeader, getEpochSwitchInfo
5. consensus/XDPoS/xdpos.go - Finalize where UpdateMasternodesFromHeader is called
6. eth/backend.go - HookPenalty implementation
7. Official comparison: /Users/anilchinchawale/github/XDCNetwork/XDPoSChain/consensus/XDPoS/engines/engine_v2/

## QUESTIONS TO ANSWER

1. What is the EXACT divergence causing the penalty mismatch? Is it:
   a) Stale smart contract state at gap blocks during bulk sync?
   b) Different preMasternodes list between GP5 and official?
   c) Different MinimumMinerBlockPerEpoch or penalty constants?
   d) Official devnet bypasses penalties somehow?

2. Should we REMOVE the bulk sync optimization entirely to achieve 100% compatibility with official XDPoSChain? What are the performance implications?

3. If we keep bulk sync, what is the correct fix to ensure UpdateMasternodesFromHeader gets correct candidates at gap blocks?

4. What is the proper fix for the V2 switch block initialization issue?

5. Are there OTHER validation bypasses we added (state root, bloom, gas used, receipt root in block_validator.go) that should also be removed if we remove bulk sync?

6. Provide a COMPLETE, PRODUCTION-READY fix that achieves byte-identical consensus with official devnet.

## CONSTRAINTS

- User explicitly demands: "if require remove bulk sync and keep it 100% compatible with geth"
- The node must sync with official devnet at 154.38.161.117:20303
- Genesis hash must remain 7dad95..629cc9 (devnet chainId 5551)
- Do NOT suggest bypasses or shortcuts - proper consensus alignment only
- If bulk sync is the root cause, recommend its complete removal