Do we want to support "in-schema" tool setup?

[nmoroze]

A small detail that came up when I was debugging error failures in #1091: do you think we should support configuring tools fully within the schema?

For example, we have a test that sets up some imaginary tools like:

...
    chip.set('tool', 'foo', 'exe', 'foo')
    chip.set('tool', 'bar', 'exe', 'foo')
    chip.set('tool', 'baz', 'exe', 'baz')
...

but there are no tool .py files. For now, check_manifest() is okay with this since the exe's are defined, and I kept this the case in my implementation of #1091. However, not having the .py file would eventually lead to an exception in run().

Supposing that these exe's are actually set correctly, do you think we should consider this legal, or should we enforce that there is a .py file for each tool used in the flow?