feat: Protect compiler directives during mutation

[dupdob]

Ensure compiler directives (#if, #pragma....) are not mutated but kept around mutated syntax nodes.

This required small changes in every mutator, as they are not responsible for removing any directive.
I took this as an opportunity to improve the design.

Also, added generation of semantic model during mutation tests to ensure mutator related unit tests are relevant.
I may have fixed some minor issues on them.

Also added some top level statement in integration test as it was a quick win

FIxes #3081, #3102

[dupdob]

I am happy to see this PR reduces the total number of lines, and it increases the number of mutations. This is probably related to the redesign of the pattern mutator that I modified so it no longer does some internal scan and reenabled orchestration of patterns in general

[Liam-Rougoor]

Thanks for your PR, @dupdob ! It's looking pretty good.

One more question, just to verify my understanding: now, all mutators should remove almost all trivia, including directives. Then, the InstrumentationEngines re-add the removed trivia, correct?

Is this process designed like this to make sure that the trivia is only added once (to prevent duplicate directive trivia, for example)?

[dupdob]

@Liam-Rougoor: short answer is yes.
Long answer is: I am not certain, it depends on what Roslyn handles as trivia and how it anchors is.
Discovering that directives were modeled as trivia was disconcerting, to say the least.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
93.4% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[dupdob]

@richardwerkman and @rouke-broersma : an important point I forgot to raise explicitly: as per @Liam-Rougoor remarks, with this PR, mutators are no longer allowed to arbitrarily mutates sub nodes.
To be more precise: mutators are no longer allowed to force the OriginalNode property of the mutant. For now, this is enforced by the general mutation orchestration that does force the value of this property.

Impacted mutators (mainly Pattern related mutators) have been modified accordingly. It turns out this behavior was induced Stryker not properly walking pattern syntax nodes, forcing the mutators to do this on their own. Stryker now walks theses nodes properly. As such, these mutators no longer need to perform discovery on their own.

This requirement allows the general orchestration to modify the original node before it being submitted for mutation, such as stripping unnecessary trivia.

[rouke-broersma]

Sounds like a good improvement, we will need to be vigilant to keep this in place.

[dupdob]

sub node mutation fails currently, as in the mutation will fail to be injected and will raise an exception, alerting the developer he/she is violating something.
I feel this is a better design, but, as there is no trivia cleaning for now (or any other change to the original node), this is actually required. So the decision could be reversed; either way, I propose will keep it as is until we have a clear benefit to choose one or the other.

I tried for days to clean trivia beforehand, but I failed to achieve a correct orchestration while doing it, so I reverted my strategy (as presented in the slack, IIRC), but kept what I feel like a cleaner design, due to the separation of concerns between orchestration and mutation

[rouke-broersma]

I don't see any reason to try to reverse this decision unless we notice that it causes other problems, as I agree that the clearer separation sounds like a good direction 👌.