Check that DoFn outputs are valid iterables when using the performance runtime type check feature

Currently, when using the performance runtime type check feature, you can yield invalid iterables from a DoFn and not have an error thrown. This is incorrect behavior.

We should throw a TypeCheckError if the iterable type is any of the following: dict, str, unicode, or bytes. 

Previously, we tried accomplishing this using an isinstance check in process_outputs of common.py, however this harms performance significantly. We should search for a cheaper way to verify that the output type is valid. One possible solution is to only do the isinstance check _if_ the flag is on.

Imported from Jira [BEAM-10782](https://issues.apache.org/jira/browse/BEAM-10782). Original Jira may contain additional context.
Reported by: saavan.