avm2: Fix Vector.<Boolean> default value in non-script-init contexts

[chabeck1]

Summary

Fixes #23317.

In Flash Player on SWF version >= 14, new Vector.<Boolean>(n, true) fills the new slots with false when called from inside a class method or constructor, but with null when called from a script initializer. Ruffle was returning null in both cases. This PR makes the non-script-init case match Flash on SWF >= 14 while leaving SWF <= 13 and the script-init case unchanged.

Approach

VectorStorage::default() for Boolean now delegates to a bool_default() helper in core/src/avm2/vector.rs that walks the AVM2 call stack from the top, finds the first bytecode method, and checks whether it is the instance initializer of a script's global class (Class::is_script_traits()). If so, the default is Value::Null (preserving the existing vector_holes behavior). Otherwise it is Value::Bool(false) for SWF >= 14 and Value::Null below.

To enable the call-stack walk inside VectorStorage, VectorStorage::new now takes &mut Activation; the change is threaded through every call site (amf.rs, object/vector_object.rs, globals/vector.rs, two globals/flash/* files). A small iter_top_down() helper was added to CallStack.

Acknowledgement of maintainer feedback

@Lord-McSweeney and @kjarosh noted on the issue that the Flash behavior here is a JIT specialization artifact, not a Vector semantic rule, and that any fix without a JIT would likely cost Vector construction performance. I understand this PR takes the "encode the JIT behavior as a runtime check" approach, which the issue thread explicitly cautioned against. I'm opening it anyway so the approach is on the record and reviewable; I fully understand if the preferred resolution is to close without merging. If there are changes that would make a variant of this acceptable (e.g. restricting the call-stack walk to SWF >= 14 paths only, or gating behind a config flag), I'm happy to iterate.

Tests

• tests/tests/swfs/avm2/vector_boolean_default_constructor/ -- SWF v14, the bug case. Expected output null then false captured from Adobe debug Flash Player.
• tests/tests/swfs/avm2/vector_boolean_default_constructor_v13/ -- SWF v13 sanity check, expected output null then null. Pins the pre-existing behavior below the version threshold.
• Full regression suite: 4016 passed; 0 failed. Pre-existing vector_holes test continues to pass.

Out of scope

The Haxe-generated static-method case from the issue (where Flash returns false regardless of SWF version) is not handled here; the discriminator appears to be Haxe's bytecode pattern rather than the Vector default rule, which is a separate investigation. Noted in the issue thread.

[Lord-McSweeney]

This fix is incorrect. The Vector is not filled as false or null depending on whether the activation is in interpreter mode or not; rather, the value is read as false or null depending on whether the activation is in interpreter mode or not. A boolean Vector is always filled with null when initialized.

[Lord-McSweeney]

Was an LLM used to write any of this code?

[chabeck1]

Yes, mostly LLM-generated. Sorry for not flagging it upfront. Your review also makes clear I was on the wrong code path (fill vs read). Close or iterate, whichever you prefer.

[Lord-McSweeney]

You can close this PR if you'd like, but if you want to implement the correct fix it's fine to do so on this PR