Only use effective number of samples if neff is explicitly specified.

[kdolum]

Don't compute or use effective number of samples unless neff is given.
Always import acor but continue silently if not found. In that case, if neff is set, call to acor fails because it is not there.
Fixes #50.

[kdolum]

I tested this using sample.py with and without acor (really encor) loaded and with and without supplying neff, and it seems to work. The autocorrelation length was extremely high, so maybe it is not being properly computed or maybe this test case is too simple in some way, but given the results from acor, the sampling process performed as it should.

[kdolum]

I tested my changes in a simple enterprise data analysis. It works fine, in a certain sense. That is to say, if I supply neff it crashes on the first call to acor (from encor) with an error such as The autocorrelation time is too long relative to the variance in dimension -1893825599. But if I don't supply neff, then it works fine. Previously it gave this error regardless of whether I supplied neff. So this provides the feature that I wanted.
Someday, someone could fix acor, but I don't think we should hold up this change for that.

[vhaasteren]

I went over all the changes. All this looks fine to me