Which issue does this PR close?

• Closes panic: arrow_cast(NULL, 'FixedSizeBinary(-1)') panics with LayoutError #22226.

Rationale for this change

arrow_cast(NULL, 'FixedSizeBinary(-1)') currently parses the type string successfully and can later panic in Arrow buffer allocation with a LayoutError. DataFusion should reject that invalid Arrow type specification before planning/simplification rewrites the function into a cast.

What changes are included in this PR?

• Adds shared arrow_cast type parsing/validation after Arrow DataType parsing.
• Rejects negative FixedSizeBinary element lengths with a DataFusion execution error.
• Reuses the same validation for arrow_try_cast so return-field planning and simplify-time rewrites agree.
• Checks nested types too, so List(FixedSizeBinary(-1)) is rejected before it reaches execution.
• Adds SQL logic regressions for arrow_cast, arrow_try_cast, and nested FixedSizeBinary(-1).

Are these changes tested?

Yes.

• cargo test -p datafusion-functions arrow_cast (compiled successfully; no tests matched the filter)
• cargo test -p datafusion-sqllogictest --test sqllogictests -- arrow_typeof.slt
• cargo test -p datafusion-functions
• cargo fmt --check
• cargo clippy -p datafusion-functions --all-targets -- -D warnings
• git diff --check

Are there any user-facing changes?

Yes. Invalid FixedSizeBinary(-1) type specs passed to arrow_cast / arrow_try_cast now return a normal DataFusion execution error instead of allowing a later panic.