Align `DataFrame::fill_null` column argument with `fill_nan`

[Nagato-Yuzuru]

Follow-up from #22702

Problem

After #22702 adds fill_nan, the two sibling methods diverge:

pub fn fill_null(&self, value: ScalarValue, columns: Vec<String>) -> Result<DataFrame>
pub fn fill_nan (&self, value: ScalarValue, columns: &[&str])     -> Result<DataFrame>

fill_nan uses the ergonomic &[&str]; fill_null still uses Vec<String> (public since 2025-02). Callers write vec!["a".to_owned()] for one and &["a"] for the other.

Proposed

Converge on one convention &[&str]

Breaking change to fill_null. Needs an upgrade guide entry, and decide a breaking change or deprecate and add.

[Nagato-Yuzuru]

take

[Nagato-Yuzuru]

Hi, @Jefffrey @kosiew . Will do this as a breaking change to fill_null (Vec → &[&str] and ScalarValue -> &ScalarValue) with an upgrade-guide note, targeting the same release as fill_nan. Or we deprecate and add a new method with the aligned signature?

I lean toward breaking (cleaner, and the name's already taken so deprecation needs a new name), but happy to go with deprecate and add. Which do you prefer?