Address Service: Resolve memory issues with large queries

[braydonf]

A few ideas for optimization of memory usage for large queries:

• getAddressSummary: Create an iterator to sum the balances as we're handling the inputs and outputs, instead of needing to keep the entire inputs and outputs in memory. These results could optionally be cached by block height, to optimize future queries.
• getAddressHistory: When retrieving the address history, instead of getting all inputs and outputs and then combining them by transaction all at once. We could do that process based on chunks through a range of block heights, and stream the response, keeping the memory usage lower. This could be a new method createAddressHistoryStream. Related: combineTransactionInfo memory usage #354

[braydonf]

Using stream transforms has helped with memory issues in combination with lowering the maxOpenFiles of leveldb. Here is some test code: https://gist.github.com/braydonf/c88bbc76a8e925357d55

[braydonf]

WIP branch is here: https://github.com/braydonf/bitcore-node/tree/large-queries