/usage suggests configuring per-subagent model overrides for skills that don't expose that control (e.g. code-review)

## Description

`/usage` surfaces generic advice whenever a chunk of token usage traces back to subagent-heavy activity, e.g.:

> 60% of your usage came from subagents under "code-review"
> If this runs frequently, consider configuring its subagents with a cheaper model or tightening their prompts.

This advice is sound for things that actually expose a model-selection knob - a custom subagent type defined in `.claude/agents/*.md` (which supports a `model:` frontmatter field), or a `Workflow` script's `agent()` calls (which take an `opts.model`). But the built-in `code-review` skill is neither of those: its finder sub-agents are spawned from logic compiled directly into the Claude Code binary, with no `SKILL.md`, settings.json field, or any other user-editable surface to pin those sub-agents to a cheaper model.

So `/usage`'s suggestion here is advice that can't actually be acted on for this specific skill - a user following it will look for a config option that doesn't exist.

## Expected behavior

Either:
1. `/usage` should only surface the "configure a cheaper model" suggestion for subagent sources that actually expose a model override (custom agent types, Workflow scripts), or
2. Built-in skills like `code-review` should expose a real model-override mechanism (e.g. an argument or settings key), so the advice becomes actionable everywhere it's given.

The more actionable lever available today for `code-review` specifically is its effort level (`low`/`medium`/`high`/etc.), which does scale sub-agent fan-out/cost - `/usage` could suggest that instead when the source is a skill without model-override support.