feat: add llm-cli-gateway as optional multi-LLM orchestration add-in

[mr-k-man]

Summary

• Adds llm-cli-gateway as an optional gstack add-in via the contrib/add-tool/ pattern
• When installed, 6 skills gain a "Multi-LLM Orchestration" section with contextual MCP tool recommendations
• Adds Gemini as a third review voice, async parallel orchestration, and session continuity — complementary to existing CODEX_SECOND_OPINION / ADVERSARIAL_STEP patterns

What's included

Component	Files
Contrib structure	contrib/add-tool/README.md, contrib/add-tool/llm-gateway/* (README, tools.json, detection.sh, install.sh, uninstall.sh)
Resolver	scripts/resolvers/llm-gateway.ts — reads tools.json, emits conditional markdown per skill with host-aware CLI filtering
Preamble detection	12 lines of bash in preamble.ts — detects llm-cli-gateway binary + per-CLI availability (claude, codex, gemini)
Template integration	{{LLM_GATEWAY_CONTEXT}} added to review, investigate, plan-eng-review, plan-ceo-review, ship, retro
Tests	109 new tests (resolver, schema validation, host suppression, integration)

Design decisions

• MCP server name: llm-cli-gw (avoids collision with simonw's llm tool)
• Complementary: Does NOT replace existing Codex exec patterns — adds Gemini + async + sessions alongside them
• Host filtering: Generalized ctx.host === t.requires_cli prevents self-invocation on any host
• Async only for parallel work: review/ship use _async variants; investigate/plan/retro use sync
• Graceful degradation: Resolver returns empty string if gateway not installed or tools.json missing

Test plan

• 109 new tests pass (resolver unit, tools.json schema, host suppression, generated SKILL.md content, preamble detection)
• 335 existing gen-skill-docs tests pass
• 71 existing host-config tests pass (golden files updated)
• bun run gen:skill-docs --host all succeeds for all 8 hosts
• Manual: install llm-cli-gateway, run /review on a real diff, verify Multi-LLM section appears and tools work