fix(acir): Use types on dynamic arrays

[vezenovm]

Description

Problem*

Resolves #4356

Supercedes #4360

Summary*

An ACIR dynamic array is a pointer to flat memory. We have been treating this flat memory as a list of fields, however, this breaks if we do in fact need accurate numeric type information such as when working black box function inputs. For example for hash inputs we set up the byte array based upon the bit size. This needs to be the correct bit size or else we will get a lot of extra garbage when calling fetch_nearest_bytes on a FieldElement.

This PR attaches a list of Vec<NumericType> to the AcirDynamicArray structure. This gives us the expected output result for sha then.

We probably could restrict the AcirDynamicArray to be created only through a constructor where we check that the value_types match the supplied len in size. I left it for a follow-up as this is a quick fix but I can do it as part of this PR.

Additional Context

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [Exceptional Case] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[TomAFrench]

LGTM

Edit: gonna hold off for the regression test.

[vezenovm]

While adding the extra test I realized each of our blackbox functions does in fact only accept an array of a singular type. I have included the test anyway as it is a good starting example for testing a dynamic array that has many possible numeric types as per this comment (#4364 (comment)). I have left the dynamic array types as a vector for correctness sake. Using a singular type on the dynamic array passes right now as we only have blackbox functions that accept arrays of a singular type. Keeping a vector of the internal types allows us to future-proof this new flatten method.