Unify Yul parsing across test suites

[cameel]

Depends on #15610. Merged.

Just a test refactor in preparation for another PR.

Currently most Yul test suites use the yul::test::parse() helper, which has two overloads which do things in wildly inconsistent ways. One uses YulStack but assumes no errors and reports only the top-level AST. The other manually runs ObjectParser and does return errors and the whole Object but only runs analysis on the top-level AST ignoring the others.

On top of that neither of them really suits my needs, because they don't allow continuing the compilation after parsing. The PR replaces them with a new, more flexible, helper that returns YulStack instead. Then makes all test suites use it.

Also:

• Most suites do accept complete Yul objects thanks to these helpers but then proceed to silently ignore any sub-objects. I made them report an unimplemented feature instead.
• Many suites do not actually print Yul compilation errors and just say that compilation failed since they're not testing the parser and assume correctness of input files. I added a helper for easy reporting of errors and made all suites consistently use it (except for those doing Boost tests).
• I made all the suites assume the EVM dialect. We still technically have a dialect setting but there's only one dialect and passing it to YulStack is a bit of a hassle so not having to do that simplifies things a lot.

[clonker]

I didn't find anything wrong with the PR, if you want to expose the dialect in the YulStack I'll reapprove afterwards

The base branch was changed.