Since the new geoprocessing service is run as the `mmw` user in
the Worker VM, that user must have access to AWS credentials.
Instead of mounting the developer's credentials into `/aws`, they
are now mounted into the `mmw` user's home folder.

Both `~/.aws/credentials` and `~/.aws/config` must be 644.

We add a task `run` and a helper method `geoprocess`. The `run`
task converts the input into the desired format, and `geoprocess`
communicates with the geoprocessing service and returns results.

`run` is a combination of `start` and `finish`: it checks whether
a result is cacheable and cached or not, and if so returns that.
Otherwise it runs `geoprocess`.

`geoprocess` is similar to `sjs_submit` in the sense that it is
POSTing to an endpoint. Unlike `sjs_submit`, which gets back a job
id, `geoprocess` receives the actual results and returns them.

`run` is designed to replace `start` and `finish` tasks in Celery
chains. So if a previous celery chain was:

    chain(geoprocessing.start.s(data),
          geoprocessing.finish.s(),
          mytasks.process_results.s())

It will now be:

    chain(geoprocessing.run.s(data),
          mytasks.process_results.s())

We likely do not need to use `choose_worker` anymore, since each
request is independent and can be run on any worker (in the right
colored stack). However, this probably needs some more thought,
and thus will be addressed in the separate issue #2117.

These old async operations are no longer used.

This version includes RasterGroupedCount and RasterGroupedAverage
operations, which make it sufficient for Analyze and TR-55 tasks.

We need azavea/ansible-java#27 to solve
AWS access issues with OpenJDK.