This PR limits the deep replacement of Evaluable._optimized_for_numpy1 to
the operations of _optimized_for_numpy, no longer involving _simplified. This
creates a logical separation between simplify and optimized and removes overlap
between the two passes. The latter (.optimized) still implies the former
(.simplified), but executes it strictly in a pre-processing pass, with the
subsequent optimization making only local adjustments to the simplified state.
The _optimized_for_numpy methods are modified to make these adjustments in a
nimble fashion rather than relying on the heavy machinery of _simplified.
This PR limits the deep replacement of
Evaluable._optimized_for_numpy1to the operations of_optimized_for_numpy, no longer involving_simplified. This creates a logical separation between simplify and optimized and removes overlap between the two passes. The latter (.optimized) still implies the former (.simplified), but executes it strictly in a pre-processing pass, with the subsequent optimization making only local adjustments to the simplified state. The_optimized_for_numpymethods are modified to make these adjustments in a nimble fashion rather than relying on the heavy machinery of_simplified.