Type inference works incorrectly with tuple slices

Returning a slice of the tuple doesn't alter the type inference output. For example:
 
`@beam.typehints.with_output_types(Tuple[int, str])`
`class Foo(beam.DoFn):`
`  def process(self, a):`
`    yield a, str(a)`
`pcoll = beam.Pipeline() `| beam.Create([1]) | beam.ParDo(Foo()) | beam.Map(lambda x: x[1:])
`print(pcoll.element_type)`
 
This prints Tuple[int, str], while the real type is str. At worst, this should return Any.

Imported from Jira [BEAM-12413](https://issues.apache.org/jira/browse/BEAM-12413). Original Jira may contain additional context.
Reported by: dmkozh.