feat(rules): add built-in token-efficiency rule

[jeff-r2026]

What

Adds a CLI built-in rule token-efficiency.md, deployed by teamai pull to every installed tool's rules directory via the existing builtin-rules mechanism (same path as teamai-recall).

It distills our internal Effective Token 使用规范 into the subset of clauses that actually change agent behavior at the system-prompt layer:

• Grounding over memory — check files/symbols/git via tools instead of guessing (anti-hallucination → fewer retries)
• Search before you read — grep/glob to locate, then bounded-range reads; narrow command output at the source
• Tools/scripts over model reasoning — mechanical work goes to linters/sed/CLIs, not step-by-step reasoning
• Minimal, focused output — no filler, show diffs not whole files
• Persist durable info to files, and YAGNI before writing new code

Human-only guideline actions (session isolation, /compact, /clear) are intentionally excluded — they can't be enforced through a system-prompt rule.

Design notes

• Unlike teamai-recall, token-efficiency is always deployed (not gated by the recall toggle) — it's a general guardrail.
• Auto-excluded from teamai push via EXCLUDED_RULE_NAMES and cleaned up on uninstall via BUILTIN_RULE_NAMES — no separate wiring needed.
• Docs updated in both docs/usage-guide.md and docs/usage-guide.zh-CN.md (new "Built-in rules" table).

Test Plan

• npx tsc --noEmit — no errors
• npx vitest run — 1767 passed, 0 failed (added 3 cases in builtin-rules.test.ts)
• npm run build — build success
• Real E2E: invoked deployBuiltinRules against a throwaway HOME; confirmed token-efficiency.md (53 lines) lands in .claude/rules/ alongside teamai-recall.md, and is still deployed when skipRecall: true

🤖 Generated with Claude Code