EM1DSurveyTD instantiation depends on layer depths

[leonfoks]

This is a design issue / feature request I think.

The EM1DSurveyTD requires the model depths in order to instantiate. I think the survey class should be independent of the subsurface structure, which would make more sense from a programmatic stand point. I would suggest removing the depth argument, of course there are repercussions down the line in doing so.

I'm sure this was done under the assumption that the model structure stays the same, and hence you can garner some efficiencies. In our case, we change our number of layers at every iteration, so (i think) we have to re-instantiate the Survey class for every forward model.