[DOC] Group libcudf column APIs into doxygen modules page

[davidwendt]

Proposal to organize libcudf column APIs documentation into doxygen groups that can be displayed on the generated Modules page. Using the @*group tags in the comments can generate an organized page of common functions. Proposal is to use the current include/cudf external API headers to organize into groups.

header	group
binaryop.h	BinaryOps
concatenate.hpp	Concatenate
copying.hpp	Copying
dlpack.hpp	DLPack
filling.hpp	Filling
groupby.hpp	GroupBy
hashing.hpp	Hashing
join.hpp	Join
merge.hpp	Merge
partitioning.hpp	Partition
quantiles.hpp	Quantiles
reduction.hpp	Reduction
replace.hpp	Replace
reshape.hpp	Reshape
rolling.hpp	Rolling Window
search.hpp	Search
sorting.hpp	Sorting
stream_compaction.hpp	Stream Compaction
transform.hpp	Transform
transpose.hpp	Transpose
unary.hpp	UnaryOps

These can be subgroups of a Columns API as well. This should make it easier to locate documentation without scrolling through the entire cudf namespace page.

[davidwendt]

Here is a picture of the modules page with the proposed groupings.

Clicking on a group will show just the functions for that group.

[harrism]

I think we can get even more consistency in the naming... we just need to decide if we like the gerund ("ing") names used in many libcudf headers. I like it -- it is consistent with Thrust documentation, and to some extent STL documentation. e.g. partitioning. Wonder if we should use "grouping" instead of "GroupBy".

Also, are two-word group names allowed? If so, then "Unary Operations", and "Binary Operations", not "UnaryOps".