Add metavariable(s) for purrr-like formulas

[machow]

dplyr functions like across allow you to apply an operation to each of a set of selected columns, using syntax of this form:

• across(<column selection>, <column application>)

Importantly, the column application step allows you to refer to both...

• the currently selected column
• the columns within the data

This means that siuba needs to differentiate...

• the currently selected column (the point of this issue)
• columns from the table (e.g. _.some_column)

Looking broadly at siuba's pipe and call behavior, there are many cases where this could surface. E.g. basically anywhere a lambda might surface in a call():

• fct_reorder(_.cyl, _.mpg, lambda x: -x.mean())

cf: https://github.com/machow/siuba/pull/413

[machow]

This could use a section of its own in the guide, but is now handled using the Fx object. See the across docs for details

https://siuba.org/guide/programming-across.html#basic-use

from siuba import _, across, Fx, group_by, mutate, summarize, filter, arrange
from siuba.data import mtcars

mtcars >> summarize(across(_[_.startswith("m"), _.endswith("p")], Fx.mean()))