fix(python): crash when schema contains nested fixed_size_list or extension type

[erandagan]

Closes #6106

AI Disclaimer: I have used Claude Code to help draft this PR and have manually reviewed its contents.

[github-actions]

PR Review

Good bugfix — the recursive _needs_substrait_placeholder is a clean improvement over the previous flat is_fixed_size_list check, and the test covers the reported scenario well.

Two items to consider:

P1: Top-level field metadata not checked

_needs_substrait_placeholder checks f.metadata is not None for fields inside a struct, but the caller passes field.type (not the field itself):

for field in self.ds.schema:
    if _needs_substrait_placeholder(field.type):   # only inspects the DataType

If a top-level schema field itself has residual metadata={} (e.g. a top-level extension type column after a lance round-trip), this won't be caught. If substrait also rejects schema-level fields with non-None metadata, this would still crash. Worth verifying whether this is a real concern or if substrait only rejects metadata on struct sub-fields.

P1: Unrelated uv.lock churn

The lockfile diff is ~4300 lines of noise (requires-python changed from >=3.10 to >=3.9, revision bumped to 3, upload-time fields added everywhere). This appears to be an unrelated uv lock regeneration. Consider reverting the lockfile change and submitting it separately, or confirming this is intentional.

[Xuanwo]

Thank you for this fix!