Faling to re-raise exceptions

In runners common.py `reraise_augmented` fails to re-raise exceptions in some case.

`raise type(exn), args, sys.exc_info()[2]`

The above code assumes that the Exception can be reconstructed from its arguments but this is not always true for custom exception classes. (e.g. `RPCError` in https://github.com/GoogleCloudPlatform/google-cloud-datastore/blob/master/python/googledatastore/connection.py#L226

This re-raising could first attempt to re-create the exception and if that does not fail raise otherwise raise the original exception as is.

(cc: [~robertwb])

Imported from Jira [BEAM-1901](https://issues.apache.org/jira/browse/BEAM-1901). Original Jira may contain additional context.
Reported by: altay.

Start in runners common.py at reraise_augmented and inspect how the current raise statement reconstructs exceptions from their arguments. Reproduce the failure with a custom exception such as the referenced RPCError, then verify that reconstruction still works when possible and that the original exception is preserved when it does not.

Written by the indexing model from the issue text.