Take GLB of types before and after type inference

Take greatest lower bound of types before and after type inference so that we can choose most accurate type from the type hints of inner and outer transforms.

Imported from Jira [BEAM-12872](https://issues.apache.org/jira/browse/BEAM-12872). Original Jira may contain additional context.
Reported by: heejong.

Start by reading the linked Jira issue BEAM-12872 for the missing context, then locate Apache Beam's type-inference entry points for inner and outer transforms. Compare the inferred types before and after inference and verify that the greatest lower bound selects the most accurate type; the issue is done when this behavior is covered by the relevant tests.

Written by the indexing model from the issue text.