docs: add when-to-use reference for implement-team

[benjamcalvin]

Summary

Adds plugins/implement-team/references/when-to-use.md — the final PR (5 of 5) of issue #69. This is a user-facing decision doc that compares the three implementation-lifecycle commands in this marketplace (/implement, /implement-team, /implement-cli) and documents the experimental caveats of /implement-team.

Closes #69.

Doc sections

• Intro — one short paragraph framing the three commands as the same lifecycle with different architectures and tradeoffs.
• Decision table — 8 rows covering routine/small tasks, default work, ambiguous/cross-module tasks, security-sensitive changes, multi-provider / CI / cost-optimization, batch/scripted runs, experimental/research work, and the fallback when agent-teams cannot be enabled.
• The three commands at a glance — one subsection per command with architecture, review model, strengths, tradeoffs. Descriptions are drawn from each plugin's current plugin.json and SKILL.md frontmatter.
• Experimental caveats for /implement-team — full list per the acceptance criteria: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 required at startup, Claude Code >= 2.1.32, one team per session, /resume and /rewind break in-process teammates, no nested teams, higher token cost advisory, tmux/iTerm2 split-pane caveat (default in-process mode is portable).
• Links — agent-teams docs (https://code.claude.com/docs/en/agent-teams) and the three skills in this marketplace.

Scope discipline

• Adds only the new file — no other plugin files touched.
• Plugin version 0.4.0 is left untouched (docs-only addition, and issue feat: add implement-team plugin (lifecycle re-architected around agent-teams) #69 PR 5 explicitly allows this).
• First use of the references/ convention described in AGENTS.md; no sibling example existed.

Test evidence

Automated tests

• ./validate-all.sh — Errors: 0, Warnings: 0, PASSED.
• claude plugin validate plugins/implement-team — Validation passed.

Manual verification

N/A — no runnable artifacts changed. This is a single static markdown reference file.

Review focus

• Accuracy of the decision table — does each row route a realistic task shape to the right command for the right reason?
• Accuracy of per-command summaries against the current state of plugins/implement-lifecycle, plugins/implement-team, and plugins/implement-cli.
• Completeness of the experimental caveats list against the acceptance criteria in issue feat: add implement-team plugin (lifecycle re-architected around agent-teams) #69.