Discussion about pipelines and where/how build them.

To unblock #321, I'm transferring the discussion about passes and pipelines here. My point is that we have already identified a need to improve how we handle/build pipelines.

Eric's (@schweitzpgi ) comment about it:

We could certainly make more/better use of prefabbed named pass pipelines. Certain collections of passes should be run together in certain orders.

The driver is the right place to build pipelines in general. That's a central "reason to exist" for a driver, which translates the combinatorics of command-line options into compiler actions. The runtime needs to assemble a pipeline as well because of late compilation, supporting Python, etc. Still using named pipelines and maybe a common "config" file could get us closer to a SPOT ideal. The early and late compilation steps aren't going to be completely identical because they're serving different purposes.

Start by reading issue #321 and this discussion, then inspect the driver and runtime areas that assemble early and late compilation pipelines. No files or tests are named, so first establish the existing pipeline-building entry points and how Python and late compilation differ. Done requires an agreed design for named pipelines and any shared configuration, not just a local edit.