Add PipelineDescriptor class

[rengolin]

This allows creating an entire pipeline via YAML files. Supported stages are: pass, transform, bundle and include. Includes are relative to the original yaml file.

Bundles will be moved later, as they're still used by some examples. Migration of examples will come later, too, once we're happy with the design.

See: https://github.com/llvm/lighthouse/wiki/Pipeline

[fschlimb]

What about allowing arrays/list rather than just single entries? This could be very handy for yaml files whichi define a traditional pipeline, generally more than of the same kind in a sequence.

[rengolin]

A list is just a sub-pipeline, imported via include. Recursive even.

While a short list would look ok:

  - pass: cse,canonicalize

a longer list would look better as multiple entries:

  - pass: "one-shot-bufferize{function-boundary-type-conversion=identity-layout-map bufferize-function-boundaries},drop-equivalent-buffer-results,buffer-deallocation-pipeline"

vs.

  - pass: one-shot-bufferize{function-boundary-type-conversion=identity-layout-map bufferize-function-boundaries}
  - pass: drop-equivalent-buffer-results
  - pass: buffer-deallocation-pipeline

Also, a list would need to allow different types (pass/transform/yaml/bundle) which would look even weirder (especially if one of the items is an include).