feat(megatron): visualize distributed training topology

[VincentCheungKokomo]

Summary

• capture Megatron parallel coordinates, rank layer ranges, and module catalogs for cluster-wide inspection
• add physical/logical placement, rank module filtering/grouping, step trends, and clearer memory evidence to the web UI
• align optimizer telemetry with completed training steps, flush deferred samples, and keep rank heartbeats within the stale TTL

Test plan

• .venv/bin/python -m pytest -q tests/regression/profiling/test_torch_probe_sampling.py tests/regression/ext/test_megatron_contract.py
• cd web && cargo test
• cd web && cargo clippy --all-targets -- -D warnings
• Run Qwen3-8B for 20 iterations on 8 GPUs with TP=2, PP=2, DP=2
• Verify cluster queries return all 8 ranks and every rank ends at allocator Step 20

Made with Cursor

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
tests/regression/ext/_megatron_contract.py	100.00% <100.00%> (ø)
tests/regression/ext/conftest.py	98.24% <100.00%> (+0.06%)	⬆️
tests/regression/ext/test_megatron_contract.py	100.00% <100.00%> (ø)
.../regression/profiling/test_torch_probe_sampling.py	99.66% <100.00%> (+0.01%)	⬆️
tests/unit/probing/fakes/test_fakes.py	95.85% <100.00%> (+0.08%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.