docs: link experiments archive + add analysis charts

[wei9072]

Summary

Adds a hyperlink to `experiments/` from the main READMEs (EN +
ZH) and turns the experiments archive's own README into an
analysis dashboard with four charts.

Charts added to `experiments/README.md`

1. Overview — 52 dirs / 26 paired / 11 models, broken down by
   task shape (Plan A / B / C / Round 9 / initial)
2. Plan B brownfield fix-rate matrix — for every (model,
   variant) cell, how many of the 3 planted SEC bugs remain.
   Computed from the actual archived files against the current
   rule library, so the numbers are reproducible. The headline:
   GPT-5.2 / GPT-5.3-codex / GPT-5.4-mini all show 3 → 0 (perfect
   fix) when paired with Aegis.
3. Plan C multi-module completion table — surfaces the
   anti-paralysis ROI: GPT-5.4-mini A variant abandoned the task
   (no `notifications.py`, no `tests.py`, 24k tokens spent on
   design proposals); B variant of the same model completed.
   Cycle introductions: 0/14.
4. Three Aegis ROI mechanisms as a Mermaid flowchart, plus a
   table marking which mechanisms were designed-in vs emergent.

Plus a direct lineage ASCII chart connecting each experiment
finding to the PR that fixed it (Round 8 codex → PR #9; Round 9
Go/Java → PR #11 / #12).

Repo housekeeping

The previous `experiments` PR accidentally committed
`31flashlite-amb-b` as a gitlink (submodule pointer) because
codex had created a nested `.git/` inside that round directory
during the agent run. This PR removes the nested `.git/` and
re-imports the 6 actual deliverable files as plain content.

Test plan

• Main README links resolve to `experiments/README.md`
• `git ls-files experiments/31flashlite-amb-b` returns 6 files
  (was 1 stale gitlink entry before this PR)
• Plan B chart numbers reproduced via:
  `for d in *-bf-?; do python3 aegis_validate.py "$d/auth.py" | grep -c security; done`
• Mermaid renders on GitHub
• CI green on push

🤖 Generated with Claude Code