Add keyword n_tries to maximum_likelihood and maximum_a_posteriori

[itsdfish]

This PR adds a keyword n_tries to run maximum_likelihood and maximum_a_posteriori from multiple starting points to mitigate problems with local maxima (see #2279). I added two unit tests and updated the doc strings. Please let me know if there are any edits you would like me to make.

[devmotion]

Generally, if local minima/maxima are problematic - wouldn't it be better to adjust the optimization method (if possible) or turn to global optimization methods? Or use a less naive multi-start optimization method such as MultistartOptimization.jl (https://docs.sciml.ai/Optimization/stable/optimization_packages/multistartoptimization/)? I'm a bit worried that exposing the functionality in the PR in an official API might make it easier for users to miss these potentially better alternatives.

[itsdfish]

I was not aware of MultistartOptimization. In that case, it would be better to use that and disregard the PR.

[mhauru]

Thanks @devmotion, that looks handy; Sorry @itsdfish that I wasn't aware of that when we talked about this.

We could still use the test case you wrote just to check that MultistartOptimization solves it. Or alternatively use your example in the docs. I feel like this is not going to be the last time local maxima come up, and would be handy to have some code to point at and say "you could try doing this".

[itsdfish]

@mhauru, no worries! It looks like at least 3 people were unaware of MultistartOptimization. Let me know if you would like me to reporpose this PR as a test for MultistartOptimization. I can update later this week.