fix: internal tuning and parameter transformations

[sebffischer]

Previously it was not possible to do something like the code below:

mlp = lrn("classif.mlp",
  activation = nn_relu,
  neurons = to_tune(ps(
    n_layers = p_int(lower = 1, upper = 10), latent = p_int(10, 500),
    .extra_trafo = function(x, param_set) {
      list(neurons = rep(x$latent, x$n_layers))
    })
  ),
  epochs = to_tune(upper = 1000L, internal = TRUE),
  patience = 100,
  validate = 0.3,
  ...
)

After this PR:

• the above code will work
• it is possible to pass the internal_search_space manually. This is useful in cases where we want to directly pass a search_space (without the to_tune() machinery). In such cases it was previously not possible to add an .extra_trafo to this search space AND to internal tuning (because the internal search space cannot have a parameter transformation).