fvalue and mse_model are -inf if only 1 regressor

Original Launchpad bug 440151: https://bugs.launchpad.net/statsmodels/+bug/440151
Reported by: josef-pktd (joep).

the tests for dimension of exog, endog for model fail because -if is not equal -inf.
The source I think is if exog is only 1d or one column, then the df_model = 0, which, I think, causes a zero division for fvalue and mse_model

It is necessary to fix test and to check whether the -inf result is actually correct.
I did the initial correction in the test for fvalue, but commented it out again, because we need to check first if -inf is correct.

In the example, the df_model should be 1 instead of 0.
I think we need a hasconstant indicator so we can use
the correct df if there is only 1 (non-constant) regressor.

If we correct df_model by not subtracting 1, then we won't get -inf in the
case of this ticket, and we can ignore that -inf != -inf with numpy <= 1.3

Start with the dimension tests for exog and endog and trace how df_model is computed for a single-regressor model. Check the resulting fvalue and mse_model calculations, including the proposed hasconstant distinction, and determine whether the expected result should be finite or -inf before updating the tests.