Allow users to interrupt model selection searches and save data

It'd provide some ease-of-use to allow `KeyboardInterrupt` to stop instances of `IncrementalSearchCV` and still have the desired parameters. i.e, something like this:

``` python
def perform_search(model, X, y):
try:
for iteration in range(...):
best_model = find_best(best_params, k)
# ...
except KeyboardInterrupt:
pass
finally:
return best_score, best_params, best_model
```

The motivation for filing this issue: I have a model selection search running. It's taking longer than I thought it would take because I mis-specified parameters. I don't want to stop the search without at least recording some information to inform my next search.

Start by reading the IncrementalSearchCV implementation and its current search loop, then inspect any existing tests for model-selection searches. Done means a KeyboardInterrupt can stop the search while retaining the best score, parameters, and model found so far.

Written by the indexing model from the issue text.