Fix Go SDK compilation for subtype slice arguments

[chris-koo]

Generated Go verification fails to compile when a descendant-interface
slice is passed to a function expecting a base-interface slice, for example
[]ISomeItem to []IAbstractItem.

We convert only mismatching slices passed to read-only transpiled verification
functions. The conversion preserves nil and evaluates each argument once.
We extend the shared regression model for non-optional and optional calls
and regenerate the affected parse, intermediate, and SDK goldens.

Refs #840. Following the review of #841, this PR isolates the compilation fix.

Validation

• Full pre-commit suite passes on Linux with Python 3.12, including 937 unit
  tests, doctests, metadata checks, and the README help check.
• Go fast check and live tests pass for list_of_classes.
• An additional manual Go test passes for nil, empty, and nonempty slices.
• The new regression test case reproduces the compilation failure with the
  unchanged upstream generator.
• The V3.2 SDK generated from the local codegen source passes go build ./....

[mristin]

Thanks, @chris-koo ! I looked at the code. Do I understand that correctly -- you make an in-situ copy of the list to satisfy the invariance in Go? Isn't that something that caller needs to do, not the called code?

Could we maybe discuss this change in a phone call live rather than over GitHub?

[mristin]

Superseded by #852.

[mristin]

Just a note to this PR: it created a copy on every invocation, which is potentially inefficient. #852 uses generics, analogous to what we did with implementation-specific Go snippets for AAS V3.1 so that no copies need be created.