Autocasting datatypes

When trying to get publicly available models to convert with coremltools, it would be very beneficial if there was auto type casting with promotable types and the common linear operators. The infrastructure appears to be in place to handle this with cast operators and type promotion checking.

It would seem that modifying the Operator base classes type verification in the __init__ would be sufficient (cast all the inputs to a particular operator to the promotable type if the base types aren't compatible with the operation.

Start by reading the Operator base classes and the existing cast operators and type-promotion checking infrastructure mentioned in the issue. Identify how type verification is performed during initialization, then determine the supported common linear operators and define tests that show compatible inputs are promoted and incompatible operations remain rejected.

Written by the indexing model from the issue text.