Add live progress streaming for background sub-agents

### Describe the feature or problem you'd like to solve

Feature request: Richer agent observability for background sub-agents.

### Proposed solution

When running multi-phase orchestrator agents (e.g., feature lifecycle with plan → implement → deliver → review), /tasks only shows tool call count and elapsed time. There is no visibility into what the agent is currently doing — which subtask, which phase, what file it is working on. A live activity log or structured progress stream from background agents would dramatically improve the developer experience for long-running agentic workflows.

## Proposed Solution Direction

### 1. Agent Status Events (lightweight)
Allow background agents to emit structured status events via a `report_progress` tool (analogous to `report_intent`):
```
report_progress({ phase: "implement", subtask: "OEA-1235", detail: "Writing domain tests" })
```
These events would surface in `/tasks` as a "last activity" column and in timeline via Ctrl+O.

### 2. Streaming Timeline for Background Agents
`/tasks ` could open a live tail view showing the agent's timeline events in real-time (tool calls, file edits, shell commands) — similar to how `docker logs -f` works. Press `q` to detach without stopping the agent.

### 3. Structured Progress Protocol
For orchestrator-style agents, define a progress protocol where agents can report:
- Current phase (plan/implement/deliver/review)
- Subtask progress (3/6 done)
- Active file being edited
- Last tool call summary

This could be rendered as a compact progress bar in the CLI:
```
[orchestrate-ABC-1234] Phase: implement ██████░░░░ 3/6 subtasks | Working on: Sample.java
```

### 4. /tasks --watch mode
A polling mode that auto-refreshes the tasks overview every N seconds, showing tool call count delta and last activity for each agent.

### Impact
Long-running orchestrator workflows (10-70+ min) currently feel like a black box. Any of these options would let developers stay informed without manually polling or asking the main agent to check.

Start by tracing the /tasks command and the background-agent timeline or event handling, since the issue names these entry points but no files or tests. Compare the proposed report_progress events, live tail view, structured protocol, and --watch mode before narrowing the scope. Done should be a defined progress experience with observable activity for long-running background agents.

Written by the indexing model from the issue text.