Add primitive arrays column wrappers

[altavir]

Primitive array columns are required for optimized big-data applications. It is also possible to add numerical DataFrame integration with MultiK or KMath.

[nikitinas]

Our current idea is to use Arrow as a backend for primitive types.
See #78

[altavir]

It is a great idea, but it will be worth it only in terms of interop with other platforms. For JVM-only, Arrow will give nothing new.

[nikitinas]

Arrow should give significant performance increase for JVM due to nullable values types support. Current implementation generates quite a lot of boxing/unboxing. It can be solved without Arrow, but I expect Arrow implementation to be faster. We will do performance benchmarks before implementation.

And we need to support Arrow I/O anyway.

[Jolanrensen]

I was experimenting with asList() wrappers. Maybe this could solve this long-standing issue:
master...primitive-array-value-columns
but it needs more testing.

[daniel-rusu]

I recommend considering Immutable Arrays as they have some surprising memory and performance benefits. Since they are immutable and their operations produce Immutable Arrays to maintain safety, this enables many optimizations that make them even faster and more efficient than regular primitive arrays.