Siuba

Much of this release is setting up for...

* fast methods to become drop-in replacements for their slow, reference implementation counterparts!
* SQL fixes and features for [funneljoin-py](https://github.com/machow/funneljoin-py)

Fixes

* Pandas anti_join breaking when on argument is a string (264)
* Sql now resets order by after joins (277)
* Fixed major regression for sql where filter with multiple arguments did strange and terrible things (277)
* Sql summarize correctly makes a subquery when group_by references a recently created column (278)
* Sql right_join had to switch lhs and rhs, but was not correctly handling all implications of that. Specifically how key columns are kept, and suffixes used. (279)

Features

* Fast grouped methods determine whether they can run quickly, fallback if not, along with a warning. Lambdas can be used to explicitly take the slower, reference implementation route (268).
* Fast grouped methods allow property operations that return dim 0 results, like dtype. (269)
* `-` in sql arrange now produces the correct `DESC`, rather than being a `-` operation (280)
* pandas arrange now works on grouped data. i.e. DataFrameGroupBy. (280)

QA

* a slew of tests!

Fixes

* anti_join in pandas no longer breaks when on argument is a string. E.g. `anti_join(df1, df2, on = "some_col")` 264
* fct_collapse now correctly keeps missing values (262)

Thanks to tmastny for the PR (248)!


import siuba.experimental.completer



* Fix lhs ops (235)
* Support kwargs in SQL count (234)

See issue 138. This release ensures summarize...

* validates results are scalar or length 1.
* uses a Series results underlying array, to issues around Series indexes in DataFrame construction.

This is a small release, designed to support the new siuba documentation.

**Features**

* added Call.map_replace method, which is like map_subcall but replaces subcalls with the result
* added to siuba.data: cars, cars_sql