Add option to disable scaled-int datatypes in InferDataTypes

[maltanar]

This PR adds an optional argument allow_scaledint_dtypes=True to InferDataTypes which controls whether SCALEDINT<x> datatypes will be inferred. If set to False, those will be inferred as FLOAT32 instead. A simple testcase is also included to verify this behaviorr.

Background: PR #53 introduced scaled-integer datatypes in InferDataTypes, mostly for the purposes of listing quantized values as scaled-integers in inferece cost. These datatypes may cause problems for backends downstream (like FINN) since they are incomplete e.g. they do not specify what the scale factor is to obtain the integer part.

[iksnagreb]

Thank you, merging this seems (have not done any exhaustive testing yet) to make our FINN builds pass again. However, I wonder whether allow_scaledint_dtypes=False would be the more reasonable default to restore FINN-compatibility? With defaulting to True, we still would have to refactor any transformation in FINN which does InferDataTypes as a kind of cleanup transformation and it might be enough to miss one transformation to slip in some scaledint to have a broken build...