* Let numpy methods handle integer size problems in AdvancedSubtensor1

* Don't run `local_uint_constant_indices` in C/python backend

Indices are always cast to int64 by the underlying methods.

Also don't run in specialize, to reduce number of passes. Other rewrites may introduce temporar indexing operations (such as x.shape[i]) which always default to int64, and it's useless to optimize immediately.

Co-authored-by: Jesse Grabowski <jessegrabowski@gmail.com>
Co-authored-by: Rob Zinkov <zaxtax@users.noreply.github.com>

* Add numba overload for Nonzero

* added numba backend and testsfor Nonzero

* Added numba backend for Nonzero

* Modified the tests and the dispatch for efficiency

Also removes redundant tests

Graph was being broken by Scalar/Tensor conversions that prevented fusion

This will always require a roll at the end, for a minimal gain

isinstance(..., int) does not recognize numpy.integers

Also remove maxsize logic

* fixed Einsum failing with repeated inputs

* Optimise the _ensure_not_equal function
Co-authored-by: Ricardo Vieira <28983449+ricardoV94@users.noreply.github.com>

* Fix einsum failing on repeated inputs

* Fix einsum failing with repeated inputs

* Added regression test for repeated inputs to the einsum function

* Fix for failing test
Co-authored-by: Ricardo Vieira <28983449+ricardoV94@users.noreply.github.com>

---------
Co-authored-by: Ricardo Vieira <28983449+ricardoV94@users.noreply.github.com>