Dfply

- Added docstrings to functions and classes.
- Added the `case_when` function.
- Fixed `arrange` to work with symbolic functions like `desc` in the function
call.
- Added `cumany` and `cumall` window functions.
- Added `if_else` function.
- Added `na_if` function.
- Added `percent_rank` function.
- Reorganization of decorator functions (better subclassing).

Fixed an issue with the way the decorators were structured, particularly
the make_symbolic, that would cause problems with returning Call objects
that would not evaluate properly. Hopefully this is now resolved.

The "coalesce" function was added from dplyr.

Some code was moved from base.py to the new vector.py file. The vector.py
file now contains functions that specifically perform operations on
series or numpy arrays (such as coalesce, desc, etc.). Test files have been
reorganized accordingly.

This release now introduces the make_symbolic decorator, which can wrap
functions to allow them to evaluate later. This is particularly (and perhaps
only) useful when you embed functions as arguments to other functions. For
example the summary and window functions.

The code for the symbolic handling decorators has been reworked. They now
inherit from a common class since they shared patterns in their code.

- make_symbolic decorator
- README updates
- desc() and order_series_by() functions
- re-imagining of the code for symbolic_evaluation, symbolic_reference
(the functionality remains unchanged)
- window and summary functions, along with their tests, have been moved around
to other files.

- `separate` and `unite` functions added.
- Summary functions added for series operations.
- README improved dramatically.
- Function docstrings added to more functions (still not all).

Moved unit tests into individual files that reflect the categories of the
functions/features they are testing. Some small bugs have been fixed as well.

The pipe decorator now copies the dataframe upon each chained function, along
with the `_grouped_by` attribute, if any. Before, operations with the pipe
functions were modifying the original dataframe (such as `mutate`).