We've been using string values for a long time, which have
thankfully not diverged. By wrapping them in constants we
ensure their standard use throughout.

For modeling, sometimes an endpoint depends on the output of
a preceding preparatory action. To handle cases when a gwlf-e/run
is given a job that is not ready yet, or has failed, we add
errors. They use the precondition required and precondition failed
HTTP status codes.

This takes the output of gwlf-e/prepare and runs it.
If the prepare job is not ready or has failed, returns an
error.

This endpoint is based on the internal start_gwlfe endpoint.
Currently, it does not handle modifications. These will be
added at a later date.

These are passing now.