feat: adds altair.datasets

[mattijn]

This PR builds upon the amazing work in #3631 by @dangotbanned.

It introduces a modernised dataset interface for Altair, now based on vega-datasets version 3.2.0 from the Vega organisation (link).

This will provide Altair with more than 70 datasets that can be lazily loaded without any hard dependency on a specific dataframe library (single hard dependency on narwhals 🫶)

Primary interface:

from altair.datasets import data

# Load with default engine (pandas)
cars_df = data.cars()

# Load with specific engine
cars_polars = data.cars(engine="polars")
cars_pyarrow = data.cars(engine="pyarrow")

# Get URL
cars_url = data.cars.url

# Set default engine for all datasets
data.set_default_engine("polars")
movies_df = data.movies()  # Uses polars engine

# List available datasets
available_datasets = data.list_datasets()

Expert interface

# Use Loader
from altair.datasets import Loader

load = Loader.from_backend("polars")
load("penguins")
load.url("penguins")

# Use direct functions
from altair.datasets import load, url

# Load a dataset
cars_df = load("cars", backend="polars")

# Get dataset URL
cars_url = url("cars")

Spatial data support
Spatial datasets are automatically loaded using:

• geopandas when using the pandas engine
• polars-st when using the polars engine

Example:

from altair.datasets import data

source = data.us_10m(engine="polars", layer="states")
source.st.plot().project(type="albersUsa")

Migration path
Migrating from the Python vega_datasets package (link) to the new altair.datasets is just a one-line change:

# old way (this is deprecated)
from vega_datasets import data

# new way (this is awesome)
from altair.datasets import data

No other changes are needed, everything will continue to work and much more becomes possible (at least, that is the aim).