Sample bundle: ML experiment trail (suggested types: Experiment, Model, Submission, Lesson)

# Summary

Migrates `docs/ml-agent-memory/` (16-principles, dashboard, competitions, experiments, skills) to OKF v0.1 format — adding YAML frontmatter (type, title, description, tags, timestamp), creating `index.md` in each subdirectory, and converting inline backtick references to markdown links.

Before: 0 concepts, 0 edges (OKF tool couldn't parse anything).
After: 6 concepts, 24 edges (full force-directed graph).

## Changes

- **6 existing files**: +YAML frontmatter + `'## 关联'` (cross-reference section)
- **3 new files**: `competitions/index.md`, `experiments/index.md`, `skills/index.md`
- **1 file modified**: `dashboard.md` (links converted to markdown format for graph edge parsing)

## Validation

```bash
enrichment-agent visualize --bundle docs/ml-agent-memory/ \
--out /tmp/ml-agent-memory_viz.html
```

Produces 6 concepts / 24 edges / 49698 bytes HTML.

## Why OKF matters here

The `ml-agent-memory/` directory was already organized as a bundle (directory of markdown files + cross-refs). OKF formalizes it:

- **Type-driven visualization**: each concept gets a color in the force-directed graph.
- **Index pages per directory**: `index.md` becomes a hub.
- **Cross-links as edges**: markdown links → graph edges automatically.
- **Timestamp + tags**: enable time-travel and filtering.

## Related PRs

- PR #5 (on `feature/submission-format-by-metric-skill`): adds the new `kaggle-submission-format-by-metric` skill (discovered during S6E2 AutoGluon rerun).
- OKF repo issue #60 (on `GoogleCloudPlatform/knowledge-catalog`): proposes ML-canonical `type` values (Experiment, Model, Submission, Lesson, Competition, Bundle).