Original, per-data ID messages are now at TRACE level, with coarser
(per-Task) status meessages at DEBUG.

These now reflect how adjustQuantum is actually being called.  I suspect
the original types reflect a reasonable aspiration: PipelineTask
subclasses ideally would be operating on a mapping that uses their internal
collection names instead of dataset type names, but the Connections
infrastructure doesn't provide a good way to do that translation (!),
so changing that here is both out-of-scope _and_ a lot of work.

The previous implementation tested all DatasetRefs (of the right type)
for compatibility with all quanta, which doesn't scale when the graph
is large.

This removes the fillQuanta step from _PipelineScaffolding, moving the
association of datasets with quanta into fillDataIds (now
connectDataIds) and prerequisite lookup into fillDatasetRefs (now
resolveDatasetRefs).

To do that, I added a new _QuantumScaffolding object to represent an
under-construction Quantum, and removed _DatasetScaffolding as it
didn't end up providing more than a simple dict would provide;
_DatasetScaffoldingDict was accordingly renamed and adjusted to
_DatasetDict.