Fix __index fallback/getter priority and cross-class member pointer registration

[Copilot]

Extends the __newindex setter-priority fix (Issue 242) to __index getters: a registered getter anywhere in the hierarchy now takes priority over a derived-class __index fallback. Also fixes two bugs introduced in the WIP commit.

__index getter priority (Issue 242 parity)

try_call_parent_index is split into two passes — extensible getters first, fallbacks second — mirroring the __newindex fix:

getGlobalNamespace(L)
    .beginClass<Base>("Base")
        .addProperty("x", &Base::getX, &Base::setX)
    .endClass()
    .deriveClass<Middle, Base>("Middle")
        .addIndexMetaMethod(&Middle::onIndex)   // catch-all fallback
    .endClass();

// Before: obj.x reads via Middle::onIndex instead of Base::getX
// After:  obj.x correctly calls Base::getX

Cross-class member pointer registration

push_class_property_getter/setter overloads used C both as the explicit registration class and as the deduced member-owner class. When registering a base-class member on a derived-class, deduction failed and the generic data-member overload won, producing compile errors. Fixed by splitting into independent C (registration class) and B (member-owner class) template parameters.

// Now compiles correctly — base-class member pointers on a derived registration
deriveClass<Derived, Base>("Derived")
    .addProperty("x", &Base::getX, &Base::setX)

__index fallback priority — preserved cases

The WIP reordering moved the fallback too aggressively, breaking two legitimate semantics. Restored:

• Static __index: static fallback must shadow registered static property getters (StaticIndexFallbackCanShadowExistingStaticProperty).
• allowOverridingMethods: instance fallback must precede class-table Lua methods so Lua-side overrides via __newindex take effect.

For plain instance __index (no allowOverridingMethods), fallback is still consulted only after all registered getters across the full hierarchy.

[kunitoki]

@copilot We should do the same for the __index fallback ?

[kunitoki]

@copilot we need a static_assert(std::is_same_v<C, B> || std:: derived_from <C, B>) in these methods (push_class_property_getter and push_class_property_setter) now taking C and B

[Copilot]

@copilot We should do the same for the __index fallback ?

Yes — try_call_parent_index_fallbacks now scans parent metatables for __index fallbacks in a second pass (after try_call_parent_index_extensibles has already checked all registered getters), mirroring the two-pass approach used for __newindex. This was implemented in 7294db0.