Stabilize CI: deterministic TestIssue1773, Xvfb retry hardening, Maven 429 backoff

[Copilot]

The Build jMonkeyEngine workflow was failing intermittently from three independent transient causes: a flaky screenshot test, an Xvfb crash that poisoned retries, and unretried Maven Central rate-limiting.

Changes

• Deterministic TestIssue1773 (ScenarioIssue1773.java): The emitter shape samples random torus vertices from the shared FastMath.rand, whose state leaked between the two parameterized invocations. The framework already fixes tpf via IsoTimer(60), so reseeding per invocation makes the render reproducible. The scene is unchanged, so existing reference images remain valid.

public static ScreenshotTest testIssue1773(boolean worldSpace) {
    FastMath.rand.setSeed(0); // isolate from shared RNG state leaked by prior invocations
    return screenshotTest(new BaseAppState() { ... });
}

• Screenshot retry hardening (main.yml): between attempts, stop Gradle daemons (./gradlew --stop) and restart Xvfb. Previously an X_PutImage / BadLength Xlib crash killed the test JVM and the corrupted X server state carried into the next retry, wasting attempts.

• Maven 429 backoff (main.yml): publishToSonatype now retries 3× with 30s/60s backoff. A single 429 Too Many Requests during dependency resolution previously failed the Deploy Java Snapshot job.

Note

Screenshot tests could not be executed in the authoring sandbox (the root buildscript requires the Android Gradle plugin from a blocked host), so confirmation that TestIssue1773 is stable should come from a CI run. The seed fix also covers the Android variant, which shares ScenarioIssue1773.