REF: Results.predict convert to array and adjust shape

[josef-pkt]

Results.predict
convert to ndarray, and check and adjust shape, with unit_tests

closes #1032
supersedes #1178

I tried to let pandas through, but it doesn't help. np.dot for example returns an ndarray.
I left the commented out code for testing DataFrame in the generic tests, but it needs special handling on before return.
I don't know if we have any wrapper code for predict. I didn't see any.

Some comments on the requirements

• CountModels require 2d even if it's only a single prediction because of exog.shape[1]
• single regressor (OLS example) needs reshape to column_vector as in AR(MA)
• I'm not sure we always have model.exog as 2d, so I check also for 1d model.exog
• I didn't find self.data.k_exog or self.model.data.k_exog, so I did a model.exog.shape check directly

[josef-pkt]

@jseabold Do you know if there is any wrapper for predict, or any examples were predict returns a DataFrame with correct index?

This PR converts every exog to ndarray. subclasses need to handle extra arrays, like exposure in count models.

[coveralls]

Coverage remained the same when pulling ada08b9 on josef-pkt:fix_predict_1032 into 01fd4a2 on statsmodels:master.

[josef-pkt]

ready to merge IMO, unless there is a problem with DataFrames.