Simplify and clean up Win32 Cursor implementation

[HeikoKlare]

Summary

This pull request simplifies and cleans up the Win32-specific Cursor class (bundles/org.eclipse.swt/Eclipse SWT/win32/org/eclipse/swt/graphics/Cursor.java) and adds a corresponding set of Win32-specific unit tests.

Note: These changes were developed with the assistance of GitHub Copilot.

---

Code Simplifications & Improvements

Change	Reason
Replace double map-lookup + setHandleForZoomLevel() with HashMap.computeIfAbsent() in win32_getHandle()	The helper method contained redundant null/containsKey guards; both are collapsed into a single idiomatic call
Change win32_getHandle() return type from boxed Long to primitive long	Consistent with Font.win32_getHandle(), Image.win32_getHandle(), and Region.win32_getHandle(); eliminates unnecessary boxing overhead; all call sites already used long
Fix latent equality bug in equals(): Long wrapper objects compared with ==	Cursor handles are OS memory addresses far outside the JVM integer cache range [-128, 127], so == on Long could silently return false for equal handles. With the primitive return type, == on long is now correct
Use Java 16 instanceof pattern matching in equals()	Eliminates the redundant explicit cast after the type check
Remove unnecessary clone() in ImageDataWithMaskCursorHandleProvider.validateMask()	The cloned ImageData was only ever read for its width/height fields — the defensive copy was pure overhead
Remove redundant static modifier on the nested CursorHandleProvider interface	Nested interfaces are implicitly static in Java
Remove redundant final modifier on private static method getOSCursorIdFromStyle()	private static methods cannot be overridden; final is meaningless here
Remove redundant final modifier on lambda-captured local variable in destroyHandlesExcept()	The variable is already effectively final without the explicit keyword

---

New Win32-Specific Tests

A new test class CursorWin32Tests was added alongside the existing Win32 test classes, following the same patterns used in RegionWin32Tests and ImagesWin32Tests.

Test	What it covers
testDisposedCursorReturnsZeroHandle	Verifies the isDisposed() guard at the top of win32_getHandle() returns 0L after disposal
testHandleIsCachedForSameZoomLevel	Verifies the computeIfAbsent() caching — repeated calls with the same zoom return the identical cached OS handle
testImageDataCursorProducesDifferentHandlesForDifferentZoomLevels	Verifies that ImageDataCursorHandleProvider scales the source image and produces a distinct OS cursor handle per zoom level
testDestroyHandlesExceptPreservesRetainedHandle	Verifies that destroyHandlesExcept() leaves the retained zoom entry intact without marking the cursor as disposed

[github-actions]

Test Results (win32)

   30 files  +2     30 suites  +2   4m 33s ⏱️ -3s
4 657 tests +4  4 584 ✅ +4  73 💤 ±0  0 ❌ ±0 
1 191 runs  +4  1 167 ✅ +4  24 💤 ±0  0 ❌ ±0 

Results for commit 68ef2f3. ± Comparison against base commit 16e501f.

♻️ This comment has been updated with latest results.

[vogella]

The new tests look like they are not so helpful, I would not not test win32_getHandle in isolation.

[HeikoKlare]

The new tests look like they are not so helpful, I would not not test win32_getHandle in isolation.

Why do you think so? Those are actual unit tests (which SWT is mostly lacking). Some of them may not be that useful, but having core functionality of per-zoom handle provision, non-redundancy etc. regression tested sounds reasonable to me to make SWT develop less "let's hope we don't break anything, but we have sufficient time to find implicitly find regressions until the release" 🙂

[vogella]

The only useful tests seem to be the last two once to me (starting with void testImageDataCursorProducesDifferentHandlesForDifferentZoomLevels).

But if the "a cursor creates a handler" tests are useful to you, that is fine for me. Just keep an open eye for AI slope, no need to pollute our repos with unnecessary tests.

[HeikoKlare]

But if the "a cursor creates a handler" tests are useful to you, that is fine for me. Just keep an open eye for AI slope, no need to pollute our repos with unnecessary tests.

Sure, we must not accept whatever an agent proposed, so I have an eye on slop. As said, I agree that some tests are rather useless, which is why I removed them. I just kept those tests that ensure important contracts of the Win32 cursor implementation.