Rename efficient-opus command, prune stale installs, fix Antigravity re-prompt

[rlorenzo]

Summary

Modernizes the delegation command now that Fable 5 is a frontier tier, and cleans up two setup rough edges surfaced along the way. Three focused commits.

Changes

refactor: rename /efficient-opus to /efficient-orchestration

• The command was framed around Opus as the top model; with Fable now above it, that framing is stale.
• The orchestrator is now described as "your current model" and subagent tiers are described relative to it, so the command reads correctly whether you run Fable, Opus, or Sonnet.
• Concrete model names are pinned in exactly one reference ladder (Haiku < Sonnet < Opus < Fable) that the delegation guidance points at, so a new model means editing one line instead of the whole document.

feat(setup): prune stale installed commands on re-run

• install_commands now removes any destination entry that carries our <!-- source: ai-coding-setup --> marker but no longer has a matching source file (a command renamed or deleted in the repo), so stale installs clean themselves up instead of lingering. This is what handles the rename above going forward.
• Safety gates: unmarked custom commands are never touched; the existing empty-source-dir early return prevents wiping a whole category; file-mode prune is depth-1 so it will not descend into nested command dirs; skill-mode leaves the separately-installed gh skill alone because it carries no marker.
• Works for both single-file commands (Claude Code) and SKILL.md directories (Codex/Copilot).

fix(setup): stop re-prompting to set up Antigravity after selection

• The tool-selection menu is the single opt-in. The claude, codex, and copilot branches proceed directly once selected, but the antigravity branch re-asked "Setup Antigravity CLI?", so it double-prompted.
• Removed the redundant confirmation (and its skip branch) so Antigravity matches the other tools. Note: selecting Antigravity in the menu now commits to installing it, the same contract the other three already have.

Testing

• bash -n setup and shellcheck setup: clean.
• markdownlint-cli2 on changed markdown: 0 errors.
• Pre-commit hooks (shellcheck, markdownlint, BATS unit tests) passed on every commit.
• Prune logic validated against a temp scenario covering: marked-orphan (removed), marked-current (kept), unmarked-custom (kept), marked-but-nested (kept), and empty-source guard (no-op). All pass.

Not included

• The prune covers commands/skills installed via install_commands. The bin/ review-loop scripts use a separate install path and are not pruned.