[Feature Request] Execution Progress Visibility: Real-time task status, step tracking, and cost preview

## Is your feature request related to a problem?

When agents execute complex multi-step tasks (especially sub-agents), users are left staring at a loading spinner with zero visibility into:

1. **What's happening right now** — Is it searching? Writing? Waiting for an API? Stuck in a loop?
2. **How far along it is** — Step 3 of 10? 80% done? About to finish?
3. **How long it will take** — No ETA, no historical timing data
4. **Whether it's stuck** — Sub-agents can silently hang for minutes with no indication

This creates severe user anxiety, especially for expensive operations where credits are being consumed during the wait.

## Describe the solution you'd like

### 1. Step-level progress indicator
- Agents declare their execution plan upfront (e.g., 5 steps: research → outline → draft → review → output)
- Each step shows: name, status (pending/running/done/failed), elapsed time
- Progress bar: `[■■■□□] 3/5 steps (research ✓, outline ✓, draft... running 45s)`

### 2. Sub-agent status dashboard
- Parent agent sees all active sub-agents in a panel
- Each sub-agent shows: task name, current step, runtime, estimated credits consumed
- Failed sub-agents highlighted with error summary

### 3. Execution timeline
- After task completion, show a timeline view:
- When each step started/ended
- Where time was spent (API wait vs. processing vs. LLM inference)
- Total credits consumed per step

### 4. Long-running task alerts
- If a step takes >30s, show a warning
- If total runtime exceeds estimated time, show an alert
- If a sub-agent hasn't produced output in >60s, suggest checking or aborting

## Describe alternatives you've considered

- **Sub-agents self-reporting via file writes**: I built a JSONL task logger (`task_log.sh`) where sub-agents write their progress, but this only works when I remember to check the file
- **Polling sub-agent status**: Possible but wastes credits on repeated status checks

## Additional context

This is the third pillar of platform observability (along with credit transparency and error handling). Together, these three features would transform the agent execution experience from a black box into a transparent, trustworthy system.

The most successful developer tools (CI/CD pipelines, cloud deployment consoles) all provide real-time step-level progress. Agent platforms should follow the same pattern.

**Priority**: P0 — Core UX gap that affects every multi-step task execution.

The issue names no files, tests, or entry points. Begin by locating the existing agent execution and loading-state flows, then determine where progress events, sub-agent status, timeline data, and the dashboard would fit; done criteria would need to be agreed before implementation.

Written by the indexing model from the issue text.