glm Binomial loglike 0log0 error

[joep]

reported by Sol 2010-11-22

>>> sm.GLM(y.astype(float),
             sm.add_constant(XYP.astype(float), prepend=True),
             family=sm.families.Binomial()).fit()
Traceback (most recent call last):
  File "<pyshell#37>", line 3, in <module>
    family=sm.families.Binomial()).fit()
  File "c:\josef\eclipsegworkspace\statsmodels-josef-experimental-gsoc\scikits\statsmodels\glm.py", line 388, in fit
    returned a nan. This could be a boundary problem and should be reported."
ValueError: The first guess on the deviance function returned a nan. This could be a boundary problem and should be reported.

[joep (josef-pktd)]

see http://bazaar.launchpad.net/~josef-pktd/statsmodels/statsmodels-josef-experimental-gsoc/revision/2194

I added 1e-200 to all instances of a*log(b) that looked like `a` and `b` could be zero.

The test cases of Sol, with either zero successes or zero failures finish now without raising an exception.

I will attach the new test script that also includes now the zero failure case.

The only disadvantage, I can see is that we create an additional temporary array, but that might be unavoidable in any case.

Ready for review