Consistent use of Units in detector data, pixel data, and noise models

[tskisner]

Work in progress on supporting consistent units between detector data, pixel data, etc.

[tskisner]

Ok, this will need rebased after #592 is merged, since there will be conflicts, but this work now includes:

• Enforce consistent units throughout the detector, pixel, and noise classes.

• Use units when doing map-domain I/O.

• Introduce a new Unit trait type for operators and templates, and support this trait type throughout the configuration system.

• Maintain units in the noise model classes, including for detector inverse variance weights.

• Check units when assigning to DetectorData objects.

• Propagate units when duplicating and redistributing Observations.

• Do not copy internal interval containing all samples when redistributing observations.

• Scale detector data when doing arithmetic with the Combine operator

• Replace old collective debug logging with Logger.debug_rank()

Also, this does not solve the slow convergence of the gain template in the case of a "perfect" time domain model. We can work on that in the next PR.