[FEATURE] Show active model in status line when skills/subagents override the session model

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Skills can specify a model field in their YAML frontmatter (e.g., model: haiku) to run with a lighter/heavier model than the session default. This works great for cost optimization.

When a skill launches with a different model, there is an inline notification:

Skill(cc-utils:skill-updater)
  Successfully loaded skill · claude-haiku-4-5-20251001

However, this notification is inconsistent (doesn't always appear), and more importantly, the status line keeps showing the session's default model (e.g., Sonnet) while the skill is actively running on a different model (e.g., Haiku).

This means users have no real-time visibility into which model is processing their request. The only way to verify is parsing *.jsonl session logs after the fact.

Proposed Solution

Show the active model in the status line when a skill or Task-spawned agent uses a model different from the session default.

For example, if the session runs on Sonnet and a skill forces Haiku:

• Before skill: status line shows Sonnet 4.5
• During skill: status line shows Haiku 4.5 (or Sonnet 4.5 → Haiku 4.5, or ↓ Haiku 4.5)
• After skill: status line reverts to Sonnet 4.5

The indicator should only appear when the model differs from the session default (no noise when models match).

As a secondary fix, the inline launch notification (Successfully loaded skill · model-name) should be made consistent — it currently doesn't always display.

Context

Related but distinct from:

• #16385 — Addressed the opposite problem (status bar incorrectly replacing the user's model with the subagent model)
• #23722 — Status line model display shared across multiple instances

This request is about observability of model routing: the model: frontmatter feature exists and works, but there's no real-time indicator of which model is actively handling the request.