Gamma distributions

[peterstangl]

This PR is a first step to address issue #209.

• The plan is to make GeneralGammaDistributionPositive a generalization of GammaDistributionPositive with arguments a, loc, and scale. However, this would break backward compatibility. Therefore, the temporary class GeneralGammaDistributionPositiveTemporary is introduced that fulfills this purpose. It will be renamed to GeneralGammaDistributionPositive in the future.
• The new class GeneralGammaCountingProcess is introduced. It takes arguments scale_factor, counts_total, counts_signal, and background_std and replaces the functionality of the current class GeneralGammaDistributionPositive (which is kept for backward compatibility for the moment, but GeneralGammaDistributionPositiveTemporary will be renamed to GeneralGammaDistributionPositive in the future)
• The new class GammaCountingProcess is introduced as an equivalent of the current GeneralGammaCountingProcess with vanishing background uncertainty. It takes arguments scale_factor, counts_total, and counts_signal.

The inheritance is now as follows:

• GammaDistributionPositive → GammaCountingProcess → GammaUpperLimit
• GeneralGammaDistributionPositive → GeneralGammaCountingProcess → GeneralGammaUpperLimit

[DavidMStraub]

I would suggest avoiding GeneralGammaDistributionPositiveTemporary because even if you have deprecation warnings, it will not be possible for users to migrate to the new class without introducing another breaking change in the future when you rename it back. You can't avoid a breaking change, so just announce it and make a clear cut; ideally, make all arguments of the new class's constructor keyword only, to make sure non-migrated code will raise rather than silently calculating a wrong distribution.

[peterstangl]

I think since GeneralGammaDistributionPositive was not used directly in flavio so far (only as base class of GeneralGammaUpperLimit, which is used for some measurements), and it probably is also not used much outside of flavio, it should be safe to break backward compatibility there. In any case, I made the new arguments of GeneralGammaDistributionPositive keyword only such that using the old interface will raise an error.