Fixes https://github.com/narwhals-dev/narwhals/actions/runs/13968615945/job/39104706209?pr=2252

* feat(DRAFT): Add `CompliantGroupBy`

Based on (#2184 (comment))

* fix: `3.8` compat

Needed for `__init__` support

* chore: export to `_compliant`

* refactor: Implement `ArrowGroupBy`

* fix(typing): Get an agreement on variance

- Dropped `native` as it got too complex
- I've done this fake `Any` thing to make `mypy` understand in multiple places

* refactor: Implement `PandasLikeGroupBy`?

- Having a hard time working out what is going on here
- All I've changed is what the refs are named

* refactor: Implement `DaskLazyGroupBy`

Much happier with this than the `pandas` one

* chore(typing): Utilize (#2251)

Fixes https://github.com/narwhals-dev/narwhals/actions/runs/13968615945/job/39104706209?pr=2252

* refactor: `CompliantGroupBy._remap_expr_name`

- Shorted in each backend
- Added docs
- Accounts for `dask` deviation from `str`

* refactor: `CompliantGroupBy._leaf_name`

- More performant to compile a single pattern and reuse everywhere
- Gives a name to a common op
- Backend code is shorter

* refactor: `CompliantGroupBy._is_simple`

Keeps this part of `pandas` in sync, despite it doing some extra name stuff

* fix: pre `3.13` protocol support

https://results.pre-commit.ci/run/github/760058710/1742478308._GAah8QXT2i9H9TbtcMHwQ

* refactor: Move most of `CompliantGroupBy` -> `DepthTrackingGroupBy`

- `_duckdb` and `_spark_like` don't need these parts (only `_dask` does)
- Also avoids needing a `TypeVar` default, which caused some issues in https://github.com/narwhals-dev/narwhals/actions/runs/13970848097/job/39111959826?pr=2252

* refactor(DRAFT): Start simplifying, aligning lazy group bys

- These two are almost identical
- Trying to reduce them as much as possible, before moving the common parts to `nw._compliant.LazyGroupBy`

* help `mypy`

https://github.com/narwhals-dev/narwhals/actions/runs/13972033263/job/39116058196?pr=2252

* refactor: Adds `LazyGroupBy`

- Greatly simplifies what each backend needs to implement
- Avoids creating and combining intermediate lists
- Avoids performing a `not in exclude` check, where `exclude` is empty
- Identified and documented a new common method `CompliantExpr._is_multi_output_agg`

* feat(typing): `Polars*GroupBy` and `Compliant*Frame`

- Will need to make (inavriant) `CompliantExprT` to type the second part correctly
- Believe that is also causing this weird error

```
Argument of type "CompliantExprT_contra@CompliantDataFrame" cannot be assigned to
parameter "exprs" of type "CompliantExprT_contra@CompliantDataFrame" in function "select"
  Type
  "CompliantExprT_contra@CompliantDataFrame" is not assignable to type
  "CompliantExprT_contra@CompliantDataFrame"

Pylance(reportArgumentType)
```

* chore: Rename pattern, remove temp doc

* Long variable names < types & docs

* refactor: listcomp < genexpr

* chore(typing): Temp ignore for false positive

* avoid unused import

* refactor: Use `Implementation.is_pandas`