[codex] Optimize feed and route sync performance

[jcastro]

Summary

• Reduces repeated IP parsing and sorting across feed, UniFi, and Omada flows.
• Skips redundant route payload construction and rollback copies on no-op syncs.
• Trims state serialization and JSON request overhead for lower CPU, memory, and wire payload size.
• Adds regression coverage for IP sorting, UniFi no-op planning, and state tuple persistence.

Measurements

• UniFi no-op plan with 8,192 IPs: ~321 ms / ~8 MiB peak before, ~1.18 ms / ~129 KiB peak after.
• parse_ip_list with 8,192 lines: ~111 ms before, ~74 ms after.
• Omada route flatten of 256 groups x 32 networks: ~260 ms before, ~47 ms with cached destinations.
• load_feed_snapshot with 12,288 Hayahora entries: ~77 ms before skipped status work, ~37 ms after.
• State snapshot conversion with 1,000 MACs: ~0.753 ms / ~17.6 KiB peak before, ~0.001 ms / ~0.8 KiB peak after.

Validation

• git diff --check
• python3 -m compileall -q src tests
• python3 -m unittest -q (155 tests OK)

ruff and mypy could not be run because those modules are not installed in this Python environment.