blockchain_db: fix db txn ending too early

The db txn in add_block ending caused the entire overarching
batch txn to stop.
Also add a new guard class so a db txn can be stopped in the
face of exceptions.

Also use a read only db txn in init when the db itself is
read only, and do not save the max tx size in that case.

src/blockchain_db/blockchain_db.cpp

@@ -211,8 +211,6 @@ uint64_t BlockchainDB::add_block( const std::pair<block, blobdata>& blck
   if (blk.tx_hashes.size() != txs.size())
     throw std::runtime_error("Inconsistent tx/hashes sizes");
 
-  block_txn_start(false);
-
   TIME_MEASURE_START(time1);
   crypto::hash blk_hash = get_block_hash(blk);
   TIME_MEASURE_FINISH(time1);
@@ -252,8 +250,6 @@ uint64_t BlockchainDB::add_block( const std::pair<block, blobdata>& blck
 
   m_hardfork->add(blk, prev_height);
 
-  block_txn_stop();
-
   ++num_calls;
 
   return prev_height;