Improve TypeDescriptor inference of DoFn's created inside a generic PTransform

Commit https://github.com/apache/incubator-beam/commit/aa7f07fa5b22f3656d52dc9e1d4557bceb87c013 introduced the ability to infer a `TypeDescriptor` from an object created inside a concrete instance of a `PTransform` and used it to simplify `SimpleFunction` usage.

We should probably look at using the same mechanism elsewhere, such as when inferring the output type of a `ParDo`.

Imported from Jira [BEAM-324](https://issues.apache.org/jira/browse/BEAM-324). Original Jira may contain additional context.
Reported by: bchambers.

Start by reading commit aa7f07fa5b22f3656d52dc9e1d4557bceb87c013 and trace how TypeDescriptor inference works for objects created inside concrete PTransform instances. Then investigate ParDo output-type inference; done means the same mechanism is applied there, with coverage for DoFns created inside generic PTransforms.

Written by the indexing model from the issue text.