[Chore][nextjs] Slimmer LLM guidance docs

[art-alexeyenko]

Slim down skills, AGENTS.md and other LLM files in samples to limit the amount of context loaded and limit token use.

• Load skills, other info only when needed
• Apply layering approach to main AGENTS.md in templates

Rough impact:

**Default stack (S0: `CLAUDE.md` + `AGENTS.md`)** — This is the main “always-on” doc cost if both are loaded.

- **nextjs:** ~18.0 KB → ~8.2 KB (**~54% smaller**; ~4500 → ~2050 approximate tokens).
- **nextjs-app-router:** ~20.5 KB → ~9.2 KB (**~55% smaller**; ~5100 → ~2300 approximate tokens).

Roughly **half the size of a short technical article** removed from the default stack, or on the order of **8–10 typewritten pages** fewer text at ~2 KB/page (very approximate).

**Skills index (S1)** — `Skills.md` alone.

- **nextjs:** ~5.3 KB → ~3.7 KB (**~30% smaller**).
- **nextjs-app-router:** ~5.2 KB → ~3.7 KB (**~29% smaller**).

**Scaffold-style load** — Compare historical worst case (two scaffold-related skills + **monolithic** `AGENTS.md`) to the intended layered path (same skills + `AGENTS-router-specifics.md` + `AGENTS-workflows-and-boundaries.md` only).

- **nextjs:** ~19.3 KB → ~9.3 KB (**~52% smaller** than monolith + skills).
- **nextjs-app-router:** ~21.8 KB → ~10.8 KB (**~50% smaller**).

If the agent still attaches **slim** `AGENTS.md` plus both skills (~9.7 KB / ~10.8 KB), that is still well below the old monolith + skills totals.

**Full corpus (S3)** — All `.md` under `.agents/`, plus root `AGENTS.md`, `Skills.md`, `CLAUDE.md`, and `.cursor/rules/*.mdc`.

- **nextjs:** ~46.3 KB → ~48.9 KB (**~6% larger**) — expected: same text, more file overhead and README / layer preambles.
- **nextjs-app-router:** ~48.8 KB → ~51.8 KB (**~6% larger**).

[illiakovalenko]

@art-alexeyenko What’s the current status of this PR? Could you please revisit it?