ENH: added routines to convert ddp between full and SA formulations

[sglyon]

Sometimes I find it helpful to have convenient, even if not optimally efficient, routines for converting between the SA and full formulations.

This PR implements such routines.

[coveralls]

Coverage increased (+0.04%) to 92.168% when pulling 9938ff0 on sl/ddp_conversions into a82ba69 on master.

[coveralls]

Coverage increased (+0.04%) to 92.168% when pulling 9938ff0 on sl/ddp_conversions into a82ba69 on master.

[oyamad]

Maybe this should be np.full((ns, na), -np.inf)?

[oyamad]

Better to write self.R > -np.inf instead of np.isfinite(self.R)?
(self.R may contain np.inf.)

[oyamad]

A comment on terminology:
I guess the terminologies should be consistent between the code and the documentation: "product formulation" should be replaced "full formulation" in the documentation, or vice versa.

(My view is that an SA formulation is already a "full" description of the model when the sets A(s) of available actions at state s are given excluding infeasible actions.)

[sglyon]

Thanks for the review. I 100% agree on all accounts and will make the changes in a few hours when I'm back at a computer

[coveralls]

Coverage increased (+0.04%) to 92.168% when pulling 79a829c on sl/ddp_conversions into a82ba69 on master.

[oyamad]

In the test code, maybe better to assign -np.inf to some entry in R and 0 in Q.
Otherwise the code looks good to me.

[sglyon]

Good call @oyamad -- I've added these tests.

[coveralls]

Coverage increased (+0.06%) to 92.188% when pulling 181dee0 on sl/ddp_conversions into a82ba69 on master.

[mmcky]

Thanks @sglyon and @oyamad for review. This looks good to me. I will leave open one more day and then merge.

[mmcky]

Thanks @sglyon.