Cast to float in nearbyint function for quantize and choose_qparams ops (#598)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/598

Casting to float prevents calling the `double` overload of the function, which does not exist in the libc of some embedded backends.
Note that we will ask Cadence about this and if/when it gets fixed on their side, we can revert this change.

Reviewed By: manuelcandales

Differential Revision: D49878712

fbshipit-source-id: 8916ba5823a30a7c1388c1ead39b31a568b86fe3