Support exactly adding differential privacy to functions with square root irrational sensitivity

[ooovi]

As discussed in these PR comments, it may be desirable to allow adding differential privacy to functions with a sensitivity that is an irrational square root of a rational number.

This will require computing the floor of the square root of a rational number exactly, as well as providing a way to create a distribution for a DifferentialPrivacyStrategy from the square of the sensitivity instead of the sensitivity itself.

This can be worked around by rounding the sensitivity up, which will retain the privacy guarantee. We might want to analyse the loss of accuracy depending on the magnitude of the sensitivity and rational number precision.