This hack has been replaced with a multi-input Butler in ap_pipe.
ap_verify still needs it in order to simulate operations repositories.

These will likely be removed in the future, depending on how much
interaction is desired between ap_verify and the details of ap_pipe.
Unfortunately, there is no clean way to parse a dataId string unless
in a command-line task or calling one.

These are the minimal changes needed to make ap_verify work with
ap_pipe again. Optimization of pipeline_driver will happen on a
different ticket.

The new code imports metrics on exit or failure instead of after each
pipeline step, allowing us to, for the moment, query the task object
for metadata. This may need to be reworked depending on how parallel
processing is handled.

This commit makes one behavior change, in that it now tries to extract
metrics from the failed pipeline step. If this process raises an
exception, it will suppress any exception raised by the pipeline. This
is an acceptable cost for the time being.