[SPARK-57727][SQL] Fix incorrect query result due to inferAdditionalConstraints incorrectly substituting attributes with non-binary-stable collations

[jiwen624]

What changes were proposed in this pull request?

Guard the attribute-substitution cases in QueryPlanConstraints.inferAdditionalConstraints with isBinaryStable, so the transitive inference is skipped when the equated attributes have a non-binary-stable type (e.g. a non-UTF8_BINARY collation). Binary-stable types are unaffected.

This mirrors the existing guard in another rule ConstantPropagation (SPARK-55647).

Why are the changes needed?

inferAdditionalConstraints infers a predicate on b from a = b and a predicate on a by substituting a with b. This is only valid when a and b are byte-for-byte interchangeable. Under a non-binary-stable collation, a = b is a collation equality (e.g. 'hello' = 'HELLO' under UTF8_LCASE), not byte equality, so substituting into a comparison evaluated in a different collation produces a wrong constraint and silently drops rows:

CREATE TABLE t (a STRING COLLATE UTF8_LCASE, b STRING COLLATE UTF8_LCASE);
INSERT INTO t VALUES ('hello', 'HELLO');
SELECT a, b FROM t WHERE a = b AND a = 'hello' COLLATE UTF8_BINARY;

Does this PR introduce any user-facing change?

Yes, fixed incorrect query result.

How was this patch tested?

Added UT.

Was this patch authored or co-authored using generative AI tooling?

Yes. Claude Code

[jiwen624]

While this fix should be correct but it blocks pushdown of same-collation, e.g.,:

-- a, b : UTF8_LCASE ; everything same collation
... t1.a = t2.b   AND   t1.a = 'x'      -- no COLLATE override
infer: t2.b = 'x'   (all UTF8_LCASE)

I'm thinking about a second approach, probably a bit more complex though.