Add Elixir and OTP CI coverage and stabilize distributed tests

[jeregrine]

Problem

Group needs CI coverage across supported Elixir/OTP versions without running soak campaigns or benchmarks on every PR. Its local examples also need module-level concurrency without racing VM-global tracing, and distributed failures need durable diagnostics.

Fix

• Run latest-stable formatting/compiler/local checks first, followed by the six-runtime matrix and the complete upstream mix test gate, including properties, chaos and the pure Jepsen checker. Enforce warnings as errors and install Java 21 for the checker.
• Add nightly/manual/release campaigns with varied seeds and scheduler counts. Preserve upstream protocol mutation and live-checker qualification, then distribute the full six-profile Jepsen soak across independent jobs without reducing its twenty five-minute histories per profile.
• Add a separate baseline/candidate performance workflow using pinned Benchee and the distributed load/recovery benchmarks.
• Split local examples into independently scheduled subsystem modules while keeping global-tracing tests serial. Start distribution lazily, bound correctness peer schedulers, and retain fresh peers per scenario.
• Synchronize partition setup and cleanup, acknowledge monitor installation, avoid cross-VM node-name collisions, and retain seeds, histories, topology, configuration, bounded shard state and queue/buffer diagnostics before peer teardown.

Supporting information

• Rebased onto the anti-entropy changes in main. All upstream local and distributed tests are retained. Superseded timestamp-conflict probes were removed in favor of upstream protocol qualification, and mutation selectors were realigned with their intended regressions.
• The matrix covers Elixir 1.19 / OTP 26–28, Elixir 1.20 / OTP 27–28, and the latest-stable pair. Actions are current SHA-pinned releases with read-only repository permissions.
• PR failure artifacts and campaign reports are retained for 14 days; performance reports for 30 days. Only main pushes trigger CI, avoiding duplicate branch/PR matrices.
• Publishing a release reruns qualification but is not automatically gated. Manually run the workflows on the candidate ref and review performance against the previous release before publishing. A runtime-version matrix does not establish mixed Group-version wire compatibility.
• Lane budgets, prerequisites, replay commands and limitations are documented in test/README.md.