[PlasmaDev5]

Makes the headless Jolt tests (PH-F36, first slice) a real gate: JoltTest now runs on the engine test framework, passes on Windows and Linux, and is picked up by the engine repository's "Build and test" workflow.

Changes

• Tests (Tests/)
  • The custom JoltTests.exe harness becomes JoltTest on plTestFramework, the framework CoreTest and RendererTest use. Same flags, JSON report, file and line on failures and a real shutdown, so it works on Linux too.
  • 13 new sub-tests: pyramid at 30/60/144 fps, removing the base of a sleeping tower, a fast box with CCD, a layer that does not collide with itself, triggers and layers, a body made of two shapes crossing a trigger, leaving one shape of a compound trigger, two bodies in one trigger, a body deleted inside a trigger, ray and sweep hit data (including a rotated box), box overlaps away from the origin, and triggers versus query shape types.
  • Checks for known bugs use OpenDefect("PH-B14", ...). They warn and are listed at the end of the run, and -joltStrictDefects makes them fail. Today: PH-B12, PH-B13, PH-B14, PH-B15, PH-B32 and one unlisted gap (a sleeping box is not woken when the body under it is deleted).
  • PH-B20, PH-B21, PH-B43 and one PH-B32 check already passed, so they are regular checks now. Two lines that asserted the PH-B13 bug as correct are now an open-defect check.
  • jolt-tests.yml is manual only: it waited 24 hours for a self-hosted runner that does not exist on every run so far.
• Plugin runtime (two !B commits)
  • JoltJobSystem: the destructor waits for the task groups of started jobs. The framework shuts core systems down after each sub-test, destroying the job system while a worker thread was still running the last task's completion callback. On Linux that crashed about 1 run in 4.
  • -Wl,-Bsymbolic for the plugin on Linux: a host that also contains Jolt (the test executable) otherwise takes over the plugin's calls into Jolt, with its own copy of Jolt's global state.
• !B Bind Jolt cloth materials in mesh render data is cherry-picked from !B Bind cloth materials after loading GPU material data #2 because the engine's pin already depends on it. Drop it here if !B Bind cloth materials after loading GPU material data #2 lands first.

Tested locally

Run	Result
Windows Development, 13 runs	all exit 0, 58 sub-tests
Windows Debug, full suite, and a 60 s soak	pass
Linux (WSL Ubuntu), 60 runs	all exit 0
Through the engine scripts at a fresh clone of this branch, Windows	JoltTest Passed 0

Not run

• The hosted GitHub workflow, Linux Debug and Release, and the ten minute R0 soak.
• The fast box with CCD also ends above the slab without CCD, so that test does not prove CCD is what holds it.
• The engine side is a separate PR that pins this branch and depends on the open NW-F01 CI PR.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}