Check if cabal's cross-section combination of flags passed to GCC is legitimate

This tracks what I found in https://github.com/haskell/cabal/pull/5356:

Investigate whether it's actually legitimate that cabal tries to combine the `extra-libraries` etc. flags from different `executable`/`test-suite`/`benchmark` sections into one invocation when checking for the existence of these C libraries by linking an example file with GCC before the build -- there might be flags across the different sections that are mutually exculsive.

To show that it's NOT legitimate, it would suffice to find two conflicting libraries or options, stick them in two different cabal sections, and see if that breaks the build.

Start from the behavior described in PR 5356: Cabal combines extra-libraries and related flags from executable, test-suite, and benchmark sections before invoking GCC. Create two sections with conflicting libraries or options and check whether linking the example file breaks the build; document whether the combination is legitimate and the observed result.

Written by the indexing model from the issue text.