docs: remove mandatory multi-agent review workflows

[shinaoka]

Summary

Make ordinary agent work direct and proportionate, as requested by the maintainer after an excessively long implementation session.

• Remove mandatory multi-agent/cross-model final review lanes; retain a main-agent integration check against agreed requirements.
• Make delegation and independent AI review opt-in, not automatic gates.
• Stop restarting design/review cycles for corrections within an agreed design.
• Replace the audit skill's mandatory subagents, remote-only target restriction and automatic implementation/PR workflow with a bounded evidence-based audit.
• Allow small fixes to record rationale/tests in the PR rather than requiring duplicate worklogs. Unsupported review claims do not themselves require code changes.

Numerical correctness, memory/data safety, required CI, human approval and performance evidence requirements are not waived. No runtime code changes.

Context/handoff: tensor4all/tenferro-rs#1758 (comment)

Verification

• Self-reviewed the three-file diff; no subagent or cross-model review used.
• git diff --check origin/main...HEAD: passed.
• Exact CI link check: all11 rules/index.md relative links resolve.
• Optional audit prompt reference exists.