`waza dev` scores a SKILL.md file's frontmatter against heuristics and
suggests improvements. It's basically equivalent to the `sensei` CLI.
Closes #32, closes #33, closes #35

### What it does

`waza dev [skill-path]` reads a SKILL.md, scores its frontmatter for
compliance (trigger phrases, anti-triggers, routing clarity, description
length, token budget), and walks the user through improvements one at a
time. In `--auto` mode it applies all suggestions without prompting.

**Scoring levels** (Low → Medium → Medium-High → High) are determined by
the presence of:
- Trigger phrases (`USE FOR:`)
- Anti-trigger phrases (`DO NOT USE FOR:`)
- Routing clarity markers (`INVOKES:`, `**WORKFLOW SKILL**`, etc.)

### Example output

```
── Iteration 1/5 ──────────────────────────────────────────

Skill: skill-creator
Score: Low
Tokens: 4522
Description: 226 chars
Triggers: 0
Anti-triggers: 0

Issues:
  ⚠️ SKILL.md is 4522 tokens (soft limit 500)

📝 Suggested improvement (triggers):
────────────────────────────────────────
USE FOR: skill-creator, about skills, core principles, quick start, advanced features.
────────────────────────────────────────
Apply this improvement? [y/N]y

  Verified: score is now Medium

── Iteration 2/5 ──────────────────────────────────────────

Skill: skill-creator
Score: Medium
Tokens: 4546
Description: 313 chars
Triggers: 5
Anti-triggers: 0

Issues:
  ⚠️ SKILL.md is 4546 tokens (soft limit 500)

📝 Suggested improvement (anti-triggers):
────────────────────────────────────────
DO NOT USE FOR: general coding questions unrelated to skill-creator, creating new projects from scratch.
────────────────────────────────────────
Apply this improvement? [y/N]y

  Verified: score is now Medium-High

✅ Target adherence level Medium-High reached!
╔══════════════════════════════════════════════════════════════════╗
║  SENSEI SUMMARY: skill-creator                                   ║
╠══════════════════════════════════════════════════════════════════╣
║  BEFORE                          AFTER                           ║
║  ──────                          ─────                           ║
║  Score: Low                      Score: Medium-High              ║
║  Tokens: 4522                    Tokens: 4574                    ║
║  Triggers: 0                     Triggers: 5                     ║
║  Anti-triggers: 0                Anti-triggers: 2                ║
║                                                                  ║
║  TOKEN STATUS: ⚠️ Over soft limit (4574 > 500)                   ║
╚══════════════════════════════════════════════════════════════════╝
```

### Flags

| Flag | Default | Description |
|------|---------|-------------|
| `--target` | `medium-high` | Target adherence level |
| `--max-iterations` | `5` | Max improvement iterations |
| `--auto` | `false` | Apply improvements without prompting |

### Implementation

New package `cmd/waza/dev/` with the following files:

- **score.go** — `HeuristicScorer` implementing a `Scorer` interface.
Pattern-matching heuristics for triggers, anti-triggers, and routing
clarity. Validation for name format, description length, and token
budget.
- **loop.go** — The core `runDevLoop` function. Steps through
description expansion, trigger addition, anti-trigger addition, and
routing clarity. Declining a suggestion skips to the next applicable
step rather than ending the loop.
- **display.go** — Terminal output formatting.
- **prompt.go** — User prompts for confirm (y/N). Uses a shared
`bufio.Scanner` to avoid consuming stdin across multiple reads.
- **root.go** — Cobra command registration.

New package `waza/internal/skill` adds `Skill` and `Frontmatter` types
and related un/marshaling functions.

### Preparatory refactor

`internal/tokens` was extracted from `cmd/waza/tokens/internal/tokens`
so the token estimation logic can be shared between the `tokens`
subcommand and the new `dev` package.

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>