Big perf gains by converting important methods to numba

Hi QuantEcon team, I'm Saurabh, building [codeflash.ai](http://codeflash.ai/). I've contributed a few performance optimizations for QuantEcon.py previously. One of the previous feedback was to convert an optimization to numba, we took that feedback and optimized the rest of the library with numba. We have exciting results to share! - [solve_discrete_riccati_system](https://github.com/codeflash-ai/QuantEcon.py/pull/37) is faster by 544%. This fundamental improvement will be a big speedup. An [alternate implementation](https://github.com/codeflash-ai/QuantEcon.py/pull/53) that speeds Markov stationary values. - [nnash](https://github.com/codeflash-ai/QuantEcon.py/pull/35) is faster by 218%. This will speed up Nash equilibrium calculations. - [integrate_variable_trajectory](https://github.com/codeflash-ai/QuantEcon.py/pull/60) is faster by 687%! This is by removing a n^2 bottleneck. There are a few others as well that I have reviewed and approved. Please [take a look](https://github.com/codeflash-ai/QuantEcon.py/pulls?q=is%3Apr+is%3Aopen+review%3Aapproved). I would love to contribute these back to QuantEcon.py. Let me know if you have any feedback for the above optimizations? I would love to collaborate with you as I open Pull Requests for these to the main project. Thanks, Saurabh