Adds the load batch enhancement - issue #7

[gforge]

Adds 2 core functionalities init_batch() and load_batch() that can be used for initializing a data-split and then loading subsequent batches of data.

The core changes are:

• Batch loading
• Started working in multiple file structure according to issue #8
• The to_tensor can now receive specific columns before converting
• Removed the autoprint from head/tail since this behavior may be unexpected and just as easy through dataframe:head():output()

Note - sorry about the size of this PR but just ignore it until you've had time reviewing the previous ones.