[codex] release 0.6.1 Agent verification modes

[greenthree]

What changed

• add references/verification-modes.md as the complete quick / normal / full Agent verification contract
• make quick mode run fixed-seed 100-round stress during final seal, while normal mode adds one blind independent solver and full mode adds proof/reference and adversarial reviewers
• route problem creation and delivery-grade review through mode selection, escalation, isolation, evidence, and SEALED parallel handoff rules in SKILL.md
• present the three modes as a concise table in the README quick start
• bump the Python Core and both npm packages to 0.6.1, with an exact probhub-skill -> probhub@0.6.1 dependency and a complete 0.6.1 changelog entry

Why

This release turns the previously documented verification modes into executable Agent Skill behavior while keeping Python Core platform-neutral. It preserves the checkpoint/seal/generation parallel workflow: a completed problem can hand off as SEALED without waiting for other problems, and the contest still publishes once through the final multi-ID build.

Validation

• python -m probhub --version and node bin/probhub.js --version -> 0.6.1
• python -m unittest tests.test_npm_packages -v -> 10 passed
• npm run check -> 422 passed, 15 skipped
• npm run pack:check -> probhub=74, probhub-skill=5
• python -X utf8 C:\Users\13952\.codex\skills\.system\skill-creator\scripts\quick_validate.py .
• git diff --check
• local phtest: python scripts\probhub.py --json lint -> 9 problems, 0 errors
• isolated same-model forward tests covered normal-mode parallel completion, forced-fast high-risk rejection, and quick-mode 100-round stress behavior