feat: Add empty_as_null and keep_nulls options to explode

[FBruzzesi]

Description

In light of polars deprecation warnings showing up in our CI, this PR expands the explode signature to match polars' one. Adapted to all backends and backported for polars (safe for 1 column explode, let polars do polars internal on multiple columns).

---

Test cases were Claude assisted, trying to cover all four flag combos for single and multi columns including null-vs-empty same-index case, and shape-mismatch errors.

---

Merging this PR would make a lot of the jobs in #3711 happy

What type of PR is this? (check all applicable)

• 💾 Refactor
• ✨ Feature
• 🐛 Bug Fix
• 🔧 Optimization
• 📝 Documentation
• ✅ Test
• 🐳 Other

[FBruzzesi]

CI failures are related to a change in polars error message for get_column

[FBruzzesi]

Regarding the pyspark-connect, I guess we can lower coverage to 90%? (it's 94 now, target is 95)

[MarcoGorelli]

thanks @FBruzzesi !

Regarding the pyspark-connect, I guess we can lower coverage to 90%? (it's 94 now, target is 95)

yup sure

[dangotbanned]

I ported a bunch of the tests from polars for this that you can steal 😉

• https://github.com/narwhals-dev/narwhals/blob/c57e72c801b5817dc9d20262029fc6c6143c31b2/tests/plan/explode_test.py
• feat(expr-ir): Add {DataFrame,Series}.explode(empty_as_null, keep_nulls) #3347

[dangotbanned]

I've marked this as high priority (#3709 (comment)) because of the CI fix.

I obviously agree with adding the flags 😅 (#3347).
But would have suggested just passing empty_as_null=True, keep_nulls=True now to fix CI - and adding the feature as a follow-up.

I haven't reviewed the changes in each backend, but if @MarcoGorelli has and is happy - then maybe we just merge?

[MarcoGorelli]

that's a good point thanks! sure, let's do that, that'll make this easier to revert if necessary