updated markov() to add tranpose keyword + default warning

[murrayrm]

This PR partially addresses issue #395 by adding a transpose keyword to the markov function that allows transposed inputs. In the current release, the markov function is not consistent with the normal time-series convention (see discussion in #395) so a warning message is generated if transpose is not set explicitly to True.

These files will need to be updated again in the 0.9.0 release so that the default for the transpose keyword changes from True to False.

[coveralls]

Coverage decreased (-0.03%) to 86.328% when pulling 438c178 on murrayrm:markov-inconsistency into 120a926 on python-control:master.

[bnavigator]

This PR itself LGTM. But are we sure that the algorithm actually works for real data? The data used in the test is worthless to determine any markov parameters. (#395 (comment))

[murrayrm]

Good point on whether the calculation was right. Looks like it wasn't...

In this latest push, I have:

• Updated the markov() algorithm to compute the correct coefficients (the old code is there for comparison, along with a new algorithm + reference).
• Reverted to the original markov_test.py (to make sure tests are backward compatible) and moved all of the "enhanced" testing into markov_test_array.py.
• Added a new unit test in markov_test_array.py to make sure the calculation is correct.