Implement generalized levelized cost (GLC) calculation and plots

[khaeru]

At today's MESSAGE meeting, @th-hara shared work done on generalized levelized cost (GLC) as a helpful tool for diagnosing MESSAGE model behaviour.

To me (at least) this includes multiple valuable innovations that could be integrated with message_ix_models.report:

1. Implement calculation of GLC, including the portion from the dual (marginal values) of constraints, following Takuya's implementation.

   The output of this could be a separate data flow (to (a) CSV or other file(s)). As an optional extra step, it could be converted into IAMC data structure and appended to the 'main' IAMC data flow.

2. Implement “active constraint markers” in plots of 'ACT', for instance as in this screenshot:

   This could be implemented in 1+ additional plot classes in .report.plot. It would require deriving a data structure that identifies, for each technology, which constraint(s) are active.

3. Implement stacked-bar "merit order" plots of active and inactive technologies, as in these screenshots:
   

   This would rely on (1), and be another plot class similar to (2). IMHO the bars should be sorted, first by the overall total GLC (per technology), lowest to highest, and then by the constraint component, highest to lowest. In this way the 'marginal' technology would always be at the right.