Design choices

I have now chosen the following to be able to explore some parameters in log and others in normal space:

1. For the user, nothing changes. I.e. boundary, projections and p0 in paramSpecs are always given in normal space.
2. loadParamSpecs.m converts boundaries and p0 where necessary.
3. In the code, each parameter will usually be stored in its respective space. This is different to before, where parameters were usually stored in normal space.
4. For simulating the model parameters are needed in normal space. This conversion is always done in evalModel.m.
5. evalModel.m also takes care of projecting parameters. This is because a projection to 0 could not be represented in log space.
6. Finally, topologicalFilteringWrapper.m takes care that the parameter point samples for the user are converted back to normal space and projected parameters are indeed projected.

What do you think, @mikolajr?