eliminate object saving when performing bulk/buffer updates

[costin]

Currently to perform efficient writes, the data is saved before being passed to ES. As Hadoop (and various libraries) perform object pooling, each entry needs to be copied otherwise the data is lost.
This causes significant memory overhead which can be alleviated by serializing early ( #3 )

[costin]

@CodeMomentum Could you expand a bit on the workflow and what do you mean by "other atlernatives"? Is this somehow related to bulk updates? If not, why not raise a separate issue?

[lorthos]

@costin You are right, I created a separate issue, sorry for the confusion

[costin]

@CodeMomentum no worries.

[costin]

Fixed in master. The object content is saved early on as byte[] so the objects themselves can be safely reused.