Recover managed child services

[kiroro123]

Summary

• Persist NetShaper-owned child services under managed_services in the recovery manifest.
• Record portal and mitmproxy process identity metadata immediately after launch, including PID, process create time, executable, argv, and ownership details.
• Teach stale recovery to verify and terminate only matching NetShaper-owned portal/mitmproxy children before restoring firewall, shaper, plugins, and sysctl state.
• Keep externally managed/verified portal listeners out of NetShaper-owned recovery state.

Why

If the NetShaper parent is killed with SIGKILL after starting the portal or mitmproxy, normal finally cleanup cannot run. Previous stale recovery restored firewall and shaping state but had no persisted child-process ownership data, so a managed service could survive the parent. This PR makes those child services recoverable while failing closed on owner or command identity mismatch.

Validation

• .venv/bin/python -m coverage run -m pytest -q -> 372 passed, 5 skipped, 16 subtests passed
• .venv/bin/python -m coverage report --fail-under=60 -> total 69.41%
• .venv/bin/python -m coverage report --include='src/netshaper/core/session_plan.py' --fail-under=95 -> 98.63%
• .venv/bin/python -m coverage report --include='src/netshaper/core/session_runner.py' --fail-under=95 -> 97.37%
• .venv/bin/python -m coverage report --include='src/netshaper/core/portal_manager.py' --fail-under=85 -> 94.33%
• .venv/bin/python -m coverage report --include='src/netshaper/core/elevated.py,src/netshaper/core/session_plan.py,src/netshaper/core/session_runner.py' --fail-under=80 -> 97.86%
• .venv/bin/python -m ruff check .
• .venv/bin/python -m mypy
• .venv/bin/python -m bandit -c pyproject.toml -r src/netshaper