Fixes #13135, ensure completude of the reactor summary but privilege failures to be last to stay human efficient

[gnodet]

Backport of #13136 to the maven-4.0.x branch.

Cherry-pick of 88f7c71, applied cleanly with no conflicts.

[gnodet-bot]

Clean backport of #13136 to maven-4.0.x. Cherry-pick applied without conflicts, both compat and impl implementations are correctly updated, and the test suite covers the new grouping behavior thoroughly.

Checked:

• Correctness: Grouping logic (0=SKIPPED/UNKNOWN, 1=SUCCESS, 2=FAILURE) is consistent across both ExecutionEventLogger implementations. The shared StringBuilder is reset via setLength(0) after each entry and the three group calls are sequential/single-threaded — no aliasing hazard.
• Behavior change — logger.error() for FAILURE lines: FAILURE module entries and BUILD FAILURE now route to error level. The IT in MavenITmng7967ArtifactHandlerLanguageTest is correctly updated to match [ERROR] BUILD FAILURE. No other IT checks for [INFO] BUILD FAILURE (confirmed via grep across the full its/ tree).
• ... separator removal: The old separator was a lossy UX approximation; the new grouped display is strictly better — skipped modules are always shown first, then successes, then failures. Tests are correctly updated to drop the "..." InOrder expectations.
• New tests: Two new test cases cover the previously untested scenario where modules are skipped alongside a build failure (one via hasExceptions() without a BuildFailure in the reactor summary, one via explicit BuildFailure). Both pass the new logger.error() assertions.
• Static analysis: No new ast-grep or semgrep findings introduced by this PR. The synchronized block flagged by ast-grep is pre-existing and correctly uses volatile.

This review was generated by an AI agent, Hermès on behalf of @gnodet.