Test the C++ generator in CI (2nd version: with more fixes included and more upstream changes)

[cpetig]

Extends #1331 by Alex

@TartanLlama would love to hear from you about these changes as well as you wrote the original test case.

There are still 26 codegen tests failing:

• import_export_func.wit: confuses function foo and namespace foo 😢
• multiversion: we don't support versioned namespaces, yet
• variants-unioning-types.wit: I guess this is import/export type incompatibility as these types are part of a world and not an interface
• resource-borrow-in-record.wit: we don't include <span> here
• 20 times TBD

[cpetig]

• lists.wit: We receive a const argument and create non-const references in lowering.
• async-trait-function.wit: todo!() in crates/cpp/src/lib.rs:1795
• resource-alias.wit: Likely still incomplete resource support
• futures.wit, streams.wit: todo!() in crates/cpp/src/lib.rs:1579
• ret-areas.wit: Multiple ret-areas? The first should be an argument buffer (rename).
• keywords.wit: Escape some more names
• simple-http.wit: Oh, we need optional to construct an expected even outside of autosar types (wrapping expected in optional gives you an unconditional default constructor)
• variants.wit: string_view/span vs string/vector confusion and expected

I guess with this amount of significant effort I will introduce an "expected codegen failure" feature and mask them for now.

[cpetig]

I just had to deploy the already existing infrastructure for expected codegen test failures, still fixing these some day would be nice 😉

[cpetig]

This should be ready to merge, as it successfully turns on CI testing of C++, but the remaining expected test failures require more work.