feat: document deployment patterns and host access implications

## Summary

Document the three deployment patterns for Shellwright, their security implications, and common use cases for each.

## Deployment Patterns

### 1. stdio mode (default)

```bash
shellwright # runs as MCP server over stdio
```

- **Access:** Full local filesystem, user permissions
- **Risk:** Shell commands run as the invoking user
- **Use case:** Local development, Claude Code integration

### 2. HTTP mode

```bash
shellwright --http --port 7498
```

- **Access:** Full access to the host system
- **Risk:** Centralized endpoint - multiple agents share one host
- **Use case:** Shared dev environments, CI/CD pipelines

### 3. Docker (isolated)

```bash
docker run -p 7498:7498 ghcr.io/dwmkerr/shellwright --http
```

- **Access:** Container filesystem only (no host access by default)
- **Risk:** Most isolated, but cannot access local files
- **Use case:** Sandboxed demos, untrusted workloads

## Documentation Needed

1. **README section** - Quick overview of the three patterns
2. **Security considerations** - What each mode can/cannot access
3. **Common use cases:**
- "I want to automate my local terminal" → stdio
- "I want a shared shell server for my team" → HTTP + auth considerations
- "I want to demo safely" → Docker
4. **Mounting volumes in Docker** - When you need specific paths available

## Example Table for Docs

| Pattern | Host Access | Isolation | Multi-agent | Setup |
|---------|-------------|-----------|-------------|-------|
| stdio | Full | None | No | Simple |
| HTTP | Full | None | Yes | Medium |
| Docker | None* | Container | Yes | Medium |

*Unless volumes mounted