Support non-GAMS models

[khaeru]

Scenario.solve() ends with the following lines:

ixmp/ixmp/core.py

Lines 1234 to 1238 in aa7b2fe

	# write to gdx, execture GAMS, read solution from gdx
	self.to_gdx(ipth, ingdx)
	run_gams(model_file, args)
	self.read_sol_from_gdx(opth, outgdx, comment,
	var_list, equ_list, check_solution)

In order to support other, non-GAMS models, the following changes could be made:

1. Extend and generalize the code in ixmp.model_settings, adding:
   • a new ixmp.model.Model abstract class,
   • an ixmp.model.GAMSModel subclass.
2. Change Scenario.to_gdx(…) to GAMSModel.write_data(scenario, …).
3. Change run_gams(…) to GAMSModel.run(scenario, …).
4. Change Scenario.read_sol_from_gdx(…) to GAMSModel.read_solution(scenario, …).

After this, users would be able to run non-GAMS models by subclassing ixmp.Model() and implementing the methods write_data(), run(), and read_sol_from_gdx().

As identified here in the discussion of #108.

[khaeru]

Here's a modified version of Dantzig's cannery/transport problem from the Julia/JuMP optimization examples: cannery.jl.txt

To check that Model has been properly abstracted, a test could be written calling this through a JuliaModel class.

[khaeru]

The branch feature/pyomo-model contains some experimental code using Pyomo.