Mark allowArbitraryExtensions as affectsProgramStructure

[jakebailey]

Per #51435 (comment) and skimming the code, this flag only controls whether or not certain diagnostics are shown (see getResolutionDiagnostic).

[jakebailey]

Excuse the force push; I wanted to rewrite the PR into the two commits so the actual changed showed the baseline change. Let me know if this looks good.

[jakebailey]

I tacked on an extra commit which sets affectsProgramStructure instead; is that the right thing? Maybe the diff provides the answer? (I am not personally very certain)

wrong

[jakebailey]

I'm a little confused about how the baseline says that semantic diagnostics aren't updated now that I've set affectsProgramStructure instead; should both affectsProgramStructure and affectsSemanticDiagnostics be set?

[jakebailey]

Just to double check, I don't need affectsSemanticDiagnostics, correct? I ran the test (now that it's correct) and the only change when this option was added was that /lib.d.ts had its diagnostics refreshed (which doesn't seem to matter?)

[weswigham]

This is probably the right level of invalidation, since while it does affect resolution diagnostics, those diagnostics in turn control which files are actually loaded into the program (since resolution results with a resolution diagnostic don't actually get traversed for further imports, iirc). So any resolutions are safe to keep between flag settings, but we may still need to refresh the program structure to find more resolutions.

[sheetalkamat]

/lib.d.ts had its diagnostics refreshed (which doesn't seem to matter?)

Thats because you have unknowingly made the resolvedModule not module which means it adds to global scope so any addition or removal will refresh diagnostics for program.