Release NetOpsBench 0.2 scalable benchmark runtime

[yyyyyt123]

Summary

This PR delivers the NetOpsBench 0.2 benchmark/runtime refactor and validated large-topology support:

• shared IncidentEngine and public simulator SDK contracts;
• native Rust Pingmesh and background traffic client agent;
• stable xlarge and fat-tree K8/K12 topology profiles and tier-aware fault placement;
• scale-aware Pingmesh, BGP, traffic, recovery, cleanup, and quarantine semantics;
• high-fanout device-down lifecycle and corrected large profile;
• removal of superseded internal paths and obsolete tests.

This is an intentional 0.2 alpha API/runtime boundary cleanup. NetOpsBench-RL remains outside this repository and consumes the public SDK.

Validation

• 744 passed, 7 deselected non-real Python tests
• Python coverage: 83.34%
• Ruff, Black, configured mypy boundary: passed
• Rust fmt, Clippy, and 29 tests: passed
• wheel import and CLI smoke outside checkout: passed
• native client image smoke: passed
• git diff --check: passed

Real infrastructure validation completed for xlarge, fat-tree K8/K12, and the corrected large profile. The final large gate verified 64 clients, 256 background flows, complete 4,032-path and 16,128 pair-port Pingmesh coverage, clean healthy observations, leaf/spine device-down behavior, recovery, and residue-free teardown.

The completed xlarge DeepSeek campaign produced 70/70 case-valid, cleanup-successful, coverage-complete cases. Agent accuracy remains a model baseline result rather than a release gate.

Known follow-up

A deliberately broader, non-CI mypy scan reports nine annotation-only issues in topology builders. They do not affect runtime behavior or the configured release gate and can be handled in a separate type-only cleanup.