Optimize fast sync: 10x speedup (4 hours → 23 minutes)

[vsnation]

Summary

• Parallel body downloads: 8 simultaneous download streams using binary search chunking, reducing body download time from ~57 min to ~16 min (3.6x)
• SQL operation skip during fast sync: Skip TxoAdd, TxoSetSpent, InsertKernel, and set_StateInputs for blocks below TxoLo — these entries are deleted by RaiseTxoLo after sync anyway. Reduces processing time from ~168 min to ~6 min (26x)
• WAL intermediate commits: Commit every 5,000 blocks in TryGoTo to prevent WAL growth to 6.9GB, keeping it under 100MB
• Definition hash skip: Skip get_Definition() for intermediate fast sync blocks; one final call at sync target validates the entire tree
• Batched cursor operations: Replace per-block cursor SQL with batched alternatives (75% reduction in cursor overhead)
• Protocol tuning: Double header pack size (8192), deeper pipeline (4x), larger body packs (50MB/16K), removed peer busy check for header requests
• SQLite tuning: WAL mode, 128MB cache, 256MB mmap, exclusive locking, tuned autocheckpoint
• Build optimizations: -O3 -march=native, disabled unused features

Results

Metric	Before	After	Speedup
Total sync time	~240 min	23.2 min	10.3x
Header download	~15 min	0.9 min (snapshot) / 15 min (fresh)	—
Body download	~57 min	15.9 min	3.6x
Block processing	~168 min	6.4 min	26x
Processing speed	~400 blk/s	10,439 blk/s	26x

Safety

• All changes are backward-compatible and do not affect consensus rules
• UTXO tree operations (RadixTree Insert/Delete/Traverse) execute for ALL blocks — state is always correct
• Only the last ~2,880 blocks (above TxoLo) process with full SQL operations
• Crash recovery preserved: transaction commits every 5,000 blocks, cursor saved every 1,000 blocks, max ~0.5s of reprocessing on restart

Test plan

• Full mainnet fast sync completed in 23.2 minutes (~3.73M blocks)
• Node operates normally after sync completion
• Restart/crash recovery verified — node resumes from last committed cursor
• Processing speed stable across all height ranges (9,000–13,000 blk/s)
• Test on Linux x86_64
• Test with different peer configurations