This was added _just_ before TaskDef was approved for deprecation on
RFC-949, and I'm sure nothing outside ctrl_mpexec is using it, so I'm
not going through a deprecation period for it.

After DM-38498 datastore records were mistakenly being attached whether
requested or not.

Apparently this was only exported publicly by the graphBuilder module,
and while that was sort of an accident (it wasn't in __all__), it also
makes sense as the only code using it, and that makes
all_dimensions_quantum_graph_builder the logical new home.

Eventually (DM-40442) these classes will be more tightly integrated,
but this piece lets us start to modernize downstream code in advance.

Like the pipeline_graph attribute of QuantumGraph, eventually
(DM-40442) we'll want this to be more tightly integrated, probably
replacing the TaskDef as the true attribute, but for now this is enough
to allow downstream code to be more forward-looking by avoiding
deprecated or soon-to-be-deprecated TaskDef interfaces.

Passing a subclass instance of Task.ConfigClass is a weird thing to do,
but it's seems harder to prohibit it than just accept it, and we have
some tests in ctrl_mpexec that depend on doing just that.

This also makes the first argument to TaskFactory.makeTask
positional-only, which is formally backwards incompatible but not
something I'm worried about in practice.

I originally hoped we could minimize how much of the
PipelineTaskConnections interface PipelineGraph would have to expose,
with the idea being that PipelineTaskConnections would be one interface
that task authors write to, and TaskNode would be a slightly different
interface that the execution system writes against, and that decoupling
those would be good for stability.  But we've always had to punch holes
to allow the task to customize its execution in various respects, and
at this point it's best to just provide access to the connections class
itself, too, even if we'd still prefer to have other TaskNode
interfaces used by the execution system where possible.

This was the last usage of BaseConnections.makeDatasetType, so we can
deprecate it now.

This also solidifies the definition of what we mean by label-range
slicing for task that lack a dependency relationship to one or both
bounds.

FieldValidationError has more useful state for diagonstics than most
other exceptions, and that's worth preserving when validation fails.

v26 came and went while this branch was sidetracked.

This provides a way to get the labels of the tasks without the
expensive (and not-always-possible) step of making a PipelineGraph.