[minor][spark] Minor bugfixes

[Dharin-shah]

This PR fixes 3 small bugs:

1. When you call asin(2.0) or acos(5.0) (values outside the valid range), PySpark is supposed to return NaN. DuckDB
   was returning NULL instead, which breaks code that expects PySpark behavior. Fixed by using a SQL workaround since
   the internal C++ layer converts NaN to NULL by design.

2. The spark.read.json() method had all the code written, but a raise NotImplementedError at the end blocked it
   from ever returning. One line removal fixed it.

3. Instead of hitting a generic NotImplementedError when encountering DuckDB union types (which PySpark doesn't
   support), it now throws a clear ContributionsAcceptedError explaining the limitation.

[evertlammerts]

Thanks! I rebased onto v1.4-andium. When the tests pass I'll do a review.