Auto merge of zcash#2993 - str4d:2977-performance-regression, r=str4d

Use block hash comparison for consistency check when loading block index

The Equihash check caused block index loading to take around 38x longer.
However, we don't need to check it directly, as the only paths to writing a
block header to disk already go through a proof-of-work check (e.g. receiving a
block over the network). By forcing the block header inside the CBlockIndex to
be re-serialized, we retain the benefits of the consistency check without the
overhead at startup.

src/txdb.cpp

@@ -314,9 +314,11 @@ bool CBlockTreeDB::LoadBlockIndexGuts()
                 pindexNew->nTx            = diskindex.nTx;
                 pindexNew->nSproutValue   = diskindex.nSproutValue;
 
+                // Consistency checks
                 auto header = pindexNew->GetBlockHeader();
-                if (!CheckEquihashSolution(&header, Params()))
-                    return error("LoadBlockIndex(): CheckEquihashSolution failed: %s", pindexNew->ToString());
+                if (header.GetHash() != pindexNew->GetBlockHash())
+                    return error("LoadBlockIndex(): block header inconsistency detected: on-disk = %s, in-memory = %s",
+                       diskindex.ToString(),  pindexNew->ToString());
                 if (!CheckProofOfWork(pindexNew->GetBlockHash(), pindexNew->nBits, Params().GetConsensus()))
                     return error("LoadBlockIndex(): CheckProofOfWork failed: %s", pindexNew->ToString());