Support for type inference for DataFrames using the DataFrame Interch…

…ange Protocol (#3114)

* Add DataFrame Interchange Protocol types from spec for type checking

* Tests shouldn't fail when pyarrow and vegafusion aren't installed, skip instead

* Compute parse_shorthand based on __dataframe__ when possible

If the pyarrow data interchange module is available and the dataset has a __dataframe__ method, infer column types based on the DataFrame Interchange Protocol.

Fall back to pandas implementation for pandas DataFrames if pyarrow is not available or pandas is older that 1.5.

* Add comment explaining why we uninstall optional dependencies

* Add pandas GitHub issue reference

* Add changelog entry

.github/workflows/build.yml

@@ -28,6 +28,14 @@ jobs:
           pip install .[dev]
           # pip install "selenium<4.3.0"
           # pip install altair_saver
+      - name: Maybe uninstall optional dependencies
+        # We uninstall pyarrow and vegafusion for one job to test that we have not
+        # accidentally introduced a hard dependency on these libraries.
+        # Uninstalling for Python 3.9 is an arbitrary choice.
+        # Also see https://github.com/altair-viz/altair/pull/3114
+        if: ${{ matrix.python-version }}=="3.9"
+        run: |
+          pip uninstall -y pyarrow vegafusion vegafusion-python-embed
       - name: Test that schema generation has no effect
         run: |
           python tools/generate_schema_wrapper.py