post-step output manipulation

Want to be able to reshape outputs on the workflow step, e.g. similar to outputEval on CommandLineTool.

Should include a way to bind outputs from the 'run' process that are not propagated to the step output object, similar to #41

Example to take a output parameter called "tooloutput" which is a record and extract "field1" and produce a workflow step output called "myoutput".

```
steps:
step1:
out:
myoutput:
source: tooloutput
valueFrom: $(self.field1)
michelsoutput:
source: foo
propagate: false
petersoutput:
valueFrom: $(outputs.michalesoutput.abc)
davidsoutput:
valueFrom: $(outputs.michalesoutput.xzy)
sashasoutput:
source: bar
type: string
valueFrom: ${return {"quux": self + outputs.michalesoutput.def }; }
outputwithdefault:
source: anoptionaloutput
default: "the default value when this returns null"
```
Optionally declare expected type on 'out' parameters to suppress checker warnings. #42

The "in", "out" and "loop" should as much as possible have the same fields / order of operations / capabilities. Although, from discussion, we probably don't need multiple sources and linkMerge on 'loop' and 'out'.

Start by reviewing the proposed step `out` fields and operation order in this issue, then read the related discussions in issues #41 and #42. Done means the workflow specification defines post-step output reshaping, binding of unpropagated run outputs, optional types, and defaults consistently with the stated examples.

Written by the indexing model from the issue text.