Direct dynamic setting or change of ccov

[nikohansen]

Similar to #346, we want to be able to set or change "$c_\text{cov}$".

Review possible ways:

• make option 'CMA_on' (and 'CMA_rankone' and 'CMA_rankmu') versatile to be read in dynamically. They are already used to (re-)set the learning rate before the update.
• call weights.finalize after self.sm.parameters(), i.e. after resetting the learning rate, (possibly conditioned to any of the multipliers have change)
• The "correct" way to change $c_\text{cov}$, say by $\alpha$, is by re-computing $c_1, c_\mu, c_c$ from $\alpha\mu_w$ where $\alpha < 1 / \mu_w$ would need to be checked carefully?
• ...

[nikohansen]

Places where related parameter settings take place

• interfaces.StatisticalModelSamplerWithZeroMeanBaseClass.parameters allows to make these settings dependent on the sampler, defines c1 and cmu.
• options_parameters.CMAParameters.set