Improve production observability for Windows Chrome profile/session usage

## Background

We are using OpenCLI in a Windows production workflow to automate logged-in browser sessions for websites such as HeyGen. It works well overall, but several production issues are hard to diagnose because profile/session state is not very observable.

## Pain points

1. It is hard to map OpenCLI profile IDs to Chrome profile directories / account hints / current logged-in site state.
2. We need a safe read-only health check that does not launch a new browser or modify the user profile.
3. Long-running browser automation needs structured JSONL progress / heartbeat output so agents can distinguish "still waiting" from "stuck".
4. Browser API failures should include structured diagnostics: status code, path, body preview, profile id, session id, current page URL, CDP attach state.
5. Windows + Git Bash + Chrome profile usage could use more docs, especially around remote debugging port and safe attach behavior.

## Suggested features

- `opencli profiles list --json`
- `opencli profile check --site --json --no-launch`
- `--jsonl` structured event output for long tasks
- Standard structured error format for browser API calls
- Windows/Chrome profile safety guide

Example profile list output:

```json
[
{
"id": "u343mvn7",
"browser": "chrome",
"profileDirectory": "Profile 31",
"sites": ["app.heygen.com"],
"lastSeenTitle": "...",
"lastSeenUrl": "...",
"accountHint": "user@example.com",
"cdpAttached": true
}
]
```

Example read-only check output:

```json
{
"ok": false,
"reason": "chrome_running_without_remote_debugging_port",
"safeToLaunch": false,
"message": "Existing Chrome instance has no CDP port; cannot attach without relaunching."
}
```

Example browser API error output:

```json
{
"ok": false,
"error": {
"type": "browser_api_http_error",
"status": 401,
"path": "/v1/project/items",
"bodyPreview": "...",
"profile": "u343mvn7",
"session": "heygen-batch-u343mvn7",
"pageUrl": "https://app.heygen.com/..."
}
}
```

## Why this matters

In production, multiple logged-in Chrome profiles are used for different accounts. Accidentally launching a new window, attaching to the wrong profile, or silently losing login state can break long-running jobs. Better profile/session introspection would make OpenCLI much safer for AI-agent production workflows.

Start by locating the existing profile/session commands, browser API error handling, and long-running task output paths in the repository. Define the scope with maintainers before changing code; done should include the agreed read-only profile check, structured JSON/JSONL diagnostics, and Windows/Chrome safety documentation with tests or examples for the supported behavior.

Written by the indexing model from the issue text.