Error checks in deployment for kNN detector

[mauicv]

torchscript doesn't like control flow so we have to use one of @torch.jit.ignore or @torch.jit.unused in order to script the kNN detector. The implication of this is that NotFitException's will not be raised by detectors that have been torch scripted, i.e. by detectors in deployment? Don't think this is something to worry about too much for now, but wonder if we should do one or both of the following:

1. Ensure this caveat is highlighted in the documentation somewhere, as I imagine users will at some point deploy unfitted detectors and get unclear error messages.
2. Open an issue for us to think about possible solutions for this, or at least keep an eye on future torch versions to see if they improve this torch script limitation.

cc: @mauicv @jklaise

Originally posted by @ascillitoe in #677 (comment)