fix(python): cap substrait below 0.85.0

[RealXuChe]

Summary

Fixes #7152.

In Substrait 0.85.0, SimpleExtensionURI was removed. The Vortex Python bindings have not been updated to use the new implementation, and they also do not set an appropriate upper bound on the dependency version.

This change caps the Python dependency at substrait<0.85.0 until the code is updated for the URN schema.

This workaround is used because it is currently difficult to make Vortex support the latest Substrait. Vortex currently consumes Substrait expressions produced by PyArrow. However, the current Arrow tree still uses Substrait v0.44.0. Until Arrow is updated, it is difficult for us to support the latest Substrait easily because of our dependency on Arrow.

See:

PyArrow uses v0.44.0: https://github.com/apache/arrow/blob/f9315d4e7fb61ac85a77c651dcee84dbfad88472/cpp/thirdparty/versions.txt#L109

We convert PyArrow Expression into Substrait type:

vortex/vortex-python/python/vortex/arrow/expression.py

Lines 47 to 50 in f200823

	def arrow_to_vortex(arrow_expression: pc.Expression, schema: pa.Schema) -> Expr:
	compat_schema = _schema_for_substrait(schema)
	substrait_object = ExtendedExpression()
	substrait_object.ParseFromString(bytes(arrow_expression.to_substrait(compat_schema))) # pyright: ignore[reportUnusedCallResult]

Testing

Did not run tests. This is a dependency metadata change only.

[RealXuChe]

I found an important mistake in this PR, and I want to correct it here.

I was misled by the way the Substrait Python packaging is split. I also only did a quick check in an existing project before opening the PR, rather than testing the install in a clean environment from scratch. That rough testing caused me to miss the problem in this PR.

Most of my comments in the issue/PR are still correct:

• The reported failure is real, caused exactly by the removal of SimpleExtensionURI
• The removal was introduced in Substrait specification 0.85.0
• PyArrow still uses an old Substrait schema, so we can't support the latest Substrait easily before PyArrow does.

However, I misunderstood how Substrait is packaged:

• The Python package substrait does not come from the substrait spec repository. It comes from the separate Python binding repo
• The Python package substrait has its own versioning and does not match the spec version
• substrait-python v0.29.0 bumped Substrait specification to the 0.85.0 schema

(substrait-protobuf and substrait share the same substrait.* import namespace, really misleading!)

So this PR was wrong. I wrote the bound as if the PyPI substrait package version matched the spec version, which it does not. The version upper bound we should actually set is substrait<0.29.0.

I have now tested the new fix in a clean virtual environment, and I confirmed that this corrected bound works.

For this issue, I will send a follow-up PR to correct the dependency version bound.