This PR makes Scalar.__dask_layers__() return self._name instead of self.key.

Quoting from the docs, the .__dask_layers__() should return the output layers:

While the DataFrame points to the output layers on which it depends directly:
>>> df.__dask_layers__()
{'filter'}

Context

I am working on a higher level cull() function (see #6438) that uses HighLevelGraph.dependencies to determine the dependency relationship between layers. Doing this work I noticed that sometimes HighLevelGraph.dependencies will refer to low level keys and not layers because of __dask_layers__() returning keys instead of names, which means HighLevelGraph.dependencies will have values that refer to non-existing keys.

What to do with Delayed?

Delayed.__dask_layers__() also returns a key. However, Delayed doesn't have a name or anything other than the key so I am not sure what to do. Does it even make sense to define __dask_layers__() when Delayed isn't a collection?

Any suggestions?