re-thinking the CAAnimation spy logic, crash during caanimation object release, support-case-#692

[markdevocht]

The problem was that the self object (delegate) was no longer valid when calling the origional method and thus crashed.
There was no way to check this validity.

---

The old approach:

1. __detox_sync_prepareDelegateIfNeeded: (lines 78-107) dynamically injects methods into the delegate's class at runtime
2. It copies the implementation from _DTXCAAnimationDelegateHelper and swizzles animationDidStop:finished:
3. When the animation completes, the swizzled method is called on the delegate - but the delegate may be deallocated

Original: 1-to-Many (Class-level swizzling)
───────────────────
UIViewAnimationBlockDelegate CLASS
(modified once, affects ALL instances)

animationDidStart: ──swizzled──▶ __detox_sync_animationDidStart:
animationDidStop: ──swizzled──▶ __detox_sync_animationDidStop:

delegate A (instance) ─▶ Animation1
delegate B (instance) ─▶ Animation2
delegate C (instance) ─▶ Animation3

• Problem: If delegate B is deallocated, Animation2's callbacks crash because the swizzled method still tries to call [self ...] on a dead object.

---

The new approach:

1. __detox_sync_setDelegate: (lines 114-138) creates a new _DTXAnimationDelegateProxy instance for each animation
2. The proxy stores a weak reference to the original delegate and becomes the animation's actual delegate
3. When the animation completes, the proxy's animationDidStop:finished: is called - it checks if the original delegate is still alive before forwarding

New: 1-to-1 (Instance-level proxy)
───────────────────
Animation1 delegate ─▶ Proxy 1 origionalDelegate (weak) ─▶ delegate A (origional)
Animation2 delegate ─▶ Proxy 2 origionalDelegate (weak) ─▶ delegate B (DEALLOC!)
Animation3 delegate ─▶ Proxy 3 origionalDelegate (weak) ─▶ delegate C (origional)

Safety: If delegate B is deallocated:
Proxy 2's originalDelegate becomes nil (weak reference)
Proxy 2's animationDidStop: checks if (delegate) → false → no call → no crash
Animations 1 and 3 are unaffected

[asafkorem]

Generally this should work, but wrote a few concerns I had while reading.. We can discuss this tomorrow and consider not replacing the delegate object and instead tracking via a CAAnimation lifecycle hook (or at least masking proxy identity/protocol checks) to avoid subtle behavior changes