Replaced RLL with faster version

[sapphrodite]

No description provided.

[jacobaubertine]

• The print() statements are still included. These print a lot of lines and seem to impact performance, so they should be removed or replaced with logging functions.
• Using the current RLL function with the 8 hazard functions on DS1, 62 of the 64 hazard function/covariate combinations are considered converged according to scipy.optimize.root(). With the new RLL, only 56 of 64 combinations converge.
• 5 of the Type III discrete Weibull combinations look like they do not converge properly when compared to the current implementation.
• All of the discrete Weibull (order 2) combinations that converged report 'nan' for log-likelihood, AIC, and BIC values in the model comparison table. The same is true for the negative binomial combinations, except they report '-inf'.

[sapphrodite]

These were both caused by an improper integration - C-SFRAT expects NANs to be returned as NANs (not INFs), and I made an off-by-one error while transcribing hazard logic. I ran the test on all 64 combinations for DS1, and couldn't find any erroneous values. I've pushed the fix - let me know if there's anything else.