Improve performance of `PartialOrd` for logical nodes

`PartialOrd` / `PartialEq` are used during planning and thus affect planning performance.

We found some improvements that are possible in
- https://github.com/apache/datafusion/pull/17438

I think `self == other` will (re) compare all entries in `self.values` and `other.values` again

we can probably avoid checking `values == other.values` here by just checking the missing field.

however, it's important to write the PartialOrd / Ord implementation so that every field is explicitly handled (addition of a new field without changing the impl should result in the compile-time error)

```suggestion
.filter(|cmp| *cmp != Ordering::Equal || self.schema == other.schema)
```

_Originally posted by @alamb in https://github.com/apache/datafusion/pull/17438#discussion_r2325757148_

Start by reading the PartialOrd and PartialEq implementations for logical nodes in PR 17438, focusing on the planning-performance discussion and the comparison of values and schema. Update the implementation so every field is explicitly handled and redundant comparisons are avoided; done means the comparison behavior is covered without relying on implicit field additions.

Written by the indexing model from the issue text.