issue in default inverse_cdf, found with simple case of Geometric

[YeungOnion]

The test output of #341 demonstrates a bug in the default implementation of inverse_cdf for a discrete distribution.

For geometric, we can implement inverse_cdf explicitly, which is generally encouraged.

[nataziel]

I'm looking into this, seems like a good place to start!