modify the regularization term and update par class

[JohnWangDataAnalyst]

1. fitting var_inv instead of var
2. modified the regularization term becomes 1/var *(param - mean)^2
3. change E-I from P to get channel EEG
4. move some methods to Abstract for Parameters and NMM classes

[Andrew-Clappison]

Some good things with simplifying the code. In particular, making progress on #107 and #108. This is also working on the task related to #115.

Is going to be a new PR focused on JR performance. In particular, improving performance with the custom_costs_JR objective function with priors. Using 1/var (or std) for the priors seemed to help.

For now, plan is to keep the options in the Par class to allow for having no priors, priors that are not fit, and priors that are fit. Might be re-discussed in the future, regarding possible alternative like setting priors with a small effect and different learning rate. May want to test various combinations first.

Might be good to discuss in more detail which things are to be simplified / what changes to the API.

May want to have cleaned up the PR into organized commits. Need to get all examples running before accepting the PR. Also, should keep the type check quality control in function arguments.