Engine interface should change to take batches of points

[pauldix]

Performance on some queries would be better if the engine took a batch of points instead of yielding each individual point. Refactor so that it takes a YieldSeries instead of YieldPoint

[elcct]

I started some work on refactoring and now it yields in series rather than individual points. This introduced new methods YieldSeries and AggregateSeries.

I only optimized Count and Timestamp aggregator for now, other aggegators fall back to use AggregatePoint instead.

I noticed that this gives about 2x faster execution of select count(value) from some_series; query and other queries seem to have improved.

Series are being collected in datastore/leveldb_shard.go and there is POINT_BATCH_SIZE constant that defines how large batch should be. This probably should be in config file

Would be great if you could take a look

[pauldix]

Would love to take a look, can you push to a branch and submit a PR?

[jvshahid]

Moved this story to 0.5.5, we should also get rid of the YieldPoint method

[jvshahid]

I think this is already done.