Optimizer misc improvements

[SolalPirelli]

In support of the bounce benchmark (source):

• Recognize loads from scala/package$ (non-null, no side effects)
• Recognize applyVoid as a specialization of apply for functions
• Stop making a list of which module loads can be elided and just use scala/*

With this, we can inline all foreach calls and we don't need an IntRef

Have you relied on LLM-based tools in this contribution?

No

How was the solution tested?

New automated tests (including the issue's reproducer, if applicable)

[mbovel]

Benchmarks started. Workflow run.

[mbovel]

Benchmarks completed. Overview.

[SolalPirelli]

kmeans has already shown bimodal behavior before. I'll look into what the optimizer does there but that can be a separate PR.