feat(server): add runtime systems status registry

[PhilipJohnBasile]

Extracted from #336 as a standalone runtime status surface.

This PR adds a provider-neutral, bounded registry and GET /v1/mtplx/systems. Components publish JSON-compatible status by name. The registry does not import, identify, configure, or control any runtime feature.

Scope:

• Thread-safe update, removal, and detached snapshot operations
• Bounded system count, per-status payload size, and validated names
• Read-only FastAPI endpoint installed by the server
• Focused tests and a reproducible overhead benchmark

Not included: feature providers, dashboard code, adaptive coordination, request capture, replay, memory management, locality, telemetry export, policy hooks, docs, or workflows.

Measurement on macOS arm64 with Python 3.13.14:

python scripts/bench_runtime_systems.py --output /tmp/runtime-systems-measurement.json
16 registered systems, 10,000 update and snapshot iterations, 1,000 HTTP GETs

update:   p50 2.208 us, p95 2.292 us, p99 2.584 us, 451,272 ops/s
snapshot: p50 19.083 us, p95 20.042 us, p99 24.208 us, 51,981 ops/s
HTTP GET: p50 339.541 us, p95 552.459 us, p99 1,019.833 us, 2,669 req/s
correctness: HTTP 200, 16 systems, revision 10,016

The HTTP figure includes FastAPI routing and JSON serialization through TestClient. The benchmark is committed at scripts/bench_runtime_systems.py.

Validation:

• python -m pytest tests/test_runtime_systems.py: 7 passed
• python -m ruff check mtplx/runtime_systems.py tests/test_runtime_systems.py scripts/bench_runtime_systems.py: passed
• python -m compileall -q mtplx/runtime_systems.py tests/test_runtime_systems.py mtplx/server/openai.py: passed
• Full create_app route smoke: HTTP 200 with the published status returned unchanged
• git diff --check: passed