fix(rust/sedona-pointcloud): projection regression

[b4l]

fixes #824 introduced in #797

[paleolimbot]

Thank you! I am guessing this is a result of the DataFusion 42 updates I attempted 😬

Can we improve the test to catch a regression like this in the future?

[jiayuasu]

A couple of suggestions before merging:

Tests

• The previous projection test covered both .las and .laz; the rewrite only covers .las. Since LAZ goes through a separate decode path, it'd be good to keep the .laz assertion (or parameterize over both).
• Assertions check column count + type of column 0, but not the actual data. The interesting failure mode in bug(rust/sedona-pointcloud): projection pushdown does not slice batches; non-prefix projections fail or return wrong data #824 was returning the right schema with the wrong column — same UInt8 shape, wrong values. An assert_eq! on a value (e.g. classification = 0 for the single point in extra.las) would lock that down.

Scope

• LasSource::new(extension, table_schema: impl Into<TableSchema>) is an unrelated public-API change. Worth pulling out into a separate PR (or dropping) so this one is purely a fix.
• The field reordering in LasSource and LasOpener is noisy in the diff without changing behaviour; reverting it would make the fix easier to cherry-pick.

[paleolimbot]

@jiayuasu I'll move your comment to a follow-up ticket to track to improve the end-to-end testing for sedona-pointcloud...there are many more situations we need to check outside of the one you highlighted and I think we can merge this as a standalone improvement while we sort that out. Exposing this in Python is probably a good first step because we can use existing readers to independently check results.