Fix drop_nulls not passing a Selector type as subset argument

[jeromepl]

The latest release (0.21.0) has a bug that breaks DataFrame#drop_nulls when used with a subset argument:

dataframe.drop_nulls(subset: "a")
# => TypeError:
#       no implicit conversion of Array into Polars::RbSelector

This comes from the following change in type signature in polars:
pola-rs/polars@8a961cd#diff-744a743b5337b77c8e26e0efabdfe8410bfd77e3d1e55ffa54eb0f29cc51e33bR1904

So we just need to apply the same change that was done in py-polars here:
pola-rs/polars@8a961cd#diff-6f7ee6a0f7f99cb158f4d68d883a118b9fab6175602e8696739d900c8b0d864fR7399

[ankane]

Hi @jeromepl, thanks for reporting and the PR! Please make the examples consistent with the Python library.

[jeromepl]

@ankane sounds good. Should I put tests for the subset argument anywhere else or drop them entirely?

[ankane]

The examples should be enough to test the subset option (bundle exec rake test:docs).

[jeromepl]

Ah right, my bad, I missed that py-polars has some examples with subset! I'll make the changes to the examples tomorrow!

[jeromepl]

@ankane the docstring examples have been updated to match py-polars

[ankane]

Great, thanks @jeromepl!

[jeromepl]

@ankane do you have any plan to cut a new release soon? This would enable us to upgrade easily to the latest version (we've been stuck on 0.19.0 because of a couple of bugs in the last releases)