Migrate the JFace performance tests off org.eclipse.test.performance

[vogella]

Applies the treatment the editor performance tests already received to the tests under org.eclipse.jface.tests.performance: they drop the performance meter, whose measurements went nowhere because the performance database has not been configured for years, and time the individual operations instead. Every test now asserts that the operation it measures actually did something, so a run can fail rather than silently report nothing, and a manual run prints min, median, 90th percentile and maximum per scenario.

The tests are also converted to JUnit 5 throughout, which they only partly were: ViewerTest declared JUnit 5 lifecycle callbacks that were never invoked for its JUnit 4 subclasses, so its measurement setup did not run at all. Dropping UIPerformanceTestRule from the tests that never touch the workbench spares them creating and building a 300 file project.

New in this change is TreeViewerExpansionTest, which covers restoring the expansion state of a tree whose elements have a recursive hashCode, the shape of an LSP document symbol. Restoring expansion looks every item up in a table keyed by TreePath, and TreePath hashes all of its segments, so such an element gets hashed once per descendant on every lookup. It reports 361056 recursive hash visits for 1884 elements, roughly 192 per element, while the same measurement with an IElementComparer supplying a cheap key reports none. That is the cost a large outline pays on every refresh, and it was not covered by any existing test.

The bundle stays excluded from the default test run, so verification times are unaffected.

[github-actions]

Test Results

   858 files  ±0     858 suites  ±0   47m 43s ⏱️ - 14m 53s
 8 157 tests ±0   7 914 ✅ +1  243 💤 ±0  0 ❌  - 1 
20 373 runs  ±0  19 719 ✅ +1  654 💤 ±0  0 ❌  - 1 

Results for commit 4d944be. ± Comparison against base commit 7bb7740.