feat(python/sedonadb): restrict DataFrame __getitem__ to single-column lookup

[jiayuasu]

Post-merge follow-up to #846 addressing @paleolimbot's review feedback.

What changes

DataFrame.__getitem__ is now strictly single-column lookup:

df["x"]    # → Expr referencing column x
df[0]      # → Expr referencing the first column
df[-1]     # → Expr referencing the last column

Previously the same method also handled df[["x","y"]] (list projection) and df[bool_expr] (filter). Those are dropped. Users go through the explicit df.select(...) and df.filter(...) entry points instead.

The motivation, mirroring Ibis's same deprecation: a single return type (Expr) lets IDEs and type checkers resolve df["x"].<method> cleanly. With the old polymorphic shape, the return type was DataFrame | Expr, which broke autocomplete on the common case.

Error paths

Key	Behavior
Unknown column name	KeyError, message lists available columns
Out-of-range int	IndexError
bool	TypeError (guarded explicitly; bool is a subclass of int in Python)
list, slice, Expr, anything else	TypeError with a message pointing at select / filter

While here — import-discipline cleanup

__getitem__, select, and filter previously had lazy in-function imports of Expr, Literal, col, and _to_expr. Per the policy in this module ("lazy imports are reserved for optional dependencies like pyarrow"), those move to module level (combined) and the method annotations switch from string forward references to the runtime-imported names.

Test plan

• 11 tests in tests/expr/test_dataframe_getitem.py cover name / positive / negative / first / out-of-range / unknown / bool / list / slice / Expr keys plus the operator-composition path on df["x"].
• All exact repr() == ... for the positive cases, per the test-policy convention locked in earlier PRs.
• Existing test_dataframe_select.py and test_dataframe_filter.py still pass (35 expr-dataframe tests total).
• pytest --doctest-modules dataframe.py passes (17 doctests including the updated __getitem__ example).
• ruff check and ruff format clean.

[Copilot]

Pull request overview

This PR narrows Python DataFrame.__getitem__ to a single-column lookup API returning Expr, aligning bracket access with predictable typing while directing projection/filtering to explicit methods.

Changes:

• Restricts df[key] to string column names and integer column positions, including negative indices.
• Adds explicit error handling for unknown names, out-of-range indices, bools, lists, slices, and Expr keys.
• Updates tests to cover the revised lookup behavior and rejected legacy shortcut forms.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/sedonadb/python/sedonadb/dataframe.py	Updates __getitem__, import placement, annotations, and related select/filter type checks.
python/sedonadb/tests/expr/test_dataframe_getitem.py	Replaces polymorphic getitem tests with single-column lookup and error-path coverage.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.