Evosax - Sep-CMA-ES

[RobertTLange]

• Reference: Ros & Hansen (2008)
• evosax Source Code: https://github.com/RobertTLange/evosax/blob/main/evosax/strategies/sep_cma_es.py
• This PR adds a CMA-ES version which imposes a diagonal structure for the estimated covariance matrix. Thereby it is a lot more memory efficient as compared to pure CMA-ES, which has do store a (d x d) matrix.
• Benchmarks and hyperparameters:

	Benchmarks	Parameters	Results (Avg)
CartPole (easy)	900 (max_iter=1000)	Link	924.3028
CartPole (hard)	600 (max_iter=1000)	Link	626.9728
MNIST	90.0 (max_iter=2000)	Link	0.9545
Brax Ant	3000 (max_iter=300)	Link	3980.9194
Waterworld	6 (max_iter=500)	Link	9.9000
Waterworld (MA)	2 (max_iter=2000)	Link	1.1875

Note: Linting doesn't pass due to import error for Open_ES - see PR #19. This has to be merged first.

[lerrytang]

There are some test errors due to renaming in evosax. I've merged the previous PR, can you rebase and solve the errors?

[RobertTLange]

Sorry for that -- Things should be fixed now.

[lerrytang]

Thank you, as always ;)