perf: skip unnecessary clone of learner's state in `resample()`

[sebffischer]

the state is overwritten anyway later.

This matters for Learners that properly implement deep cloning. For torch learners this can be expensive and should be avoided.