test: improve array_distinct test coverage and incompatibility description

[andygrove]

Which issue does this PR close?

Closes #3174.

Rationale for this change

The existing array_distinct SQL file tests only covered array<int> with 2 queries. The Incompatible support level had no description, making it unclear to users why the expression is marked incompatible.

What changes are included in this PR?

1. Expanded SQL file tests for array_distinct from 2 queries to comprehensive coverage across 7 element types:

   • INT: duplicates, empty array, NULL input, NULL elements, all-NULLs, boundary values (INT_MIN/MAX), negative values
   • BIGINT: duplicates, NULL elements, boundary values (Long.MIN/MAX)
   • STRING: duplicates, empty strings, NULL elements, empty-string-vs-NULL distinction
   • BOOLEAN: all combinations of true/false with NULLs
   • DOUBLE: duplicates, NaN deduplication, NaN+NULL, Infinity/-Infinity, negative zero
   • FLOAT: duplicates, NaN deduplication
   • DECIMAL(10,2): duplicates, NULL elements
   • Nested arrays: array<array<int>> with duplicates and NULLs

2. Added descriptive reason to Incompatible support level: "Output elements are sorted rather than preserving insertion order".

How are these changes tested?

All new tests are SQL file tests that run via CometSqlFileTestSuite with the ConfigMatrix for both dictionary-encoded and plain parquet. Tests use spark_answer_only mode since the expression is Incompatible (DataFusion sorts output elements while Spark preserves insertion order). Verified passing locally with both dictionary configurations.