Improve Traceability of Pipeline translation

Users often ask how they can reason about the pipeline translation. The Flink UI display a confusingly large graph without any trace of the original Beam pipeline:

WordCount:

!wordcount.png!

TFX:

!tfx.png!

Some aspects which make understanding these graphs hard:
* Users don't know how the Runner maps Beam to Flink concepts
* The UI is awfully slow / hangs when the pipeline is reasonable complex
* The operator names seem to use `transform.getUniqueName()` which doesn't generate readable name
* So called Chaining combines operators into a single operator which makes understanding which Beam concept belongs to which Flink concept even harder

Imported from Jira [BEAM-5859](https://issues.apache.org/jira/browse/BEAM-5859). Original Jira may contain additional context.
Reported by: mxm.