Fix session persistence, Cloudflare challenge, and empty get-my-articles

[emregulerr]

Fixes #1.

The authenticated tools returned empty results and forced repeated logins. While debugging I found several overlapping root causes in the browser layer and fixed them together.

What was wrong

1. Session cookies lost across context restarts. Medium's auth relies on session cookies (sid/xsrf). A persistent profile keeps cookies with an expiry, but Chromium drops pure session cookies on close/reopen — so the headed→headless switch and server restarts silently logged the user out.
2. Cloudflare "Just a moment…" challenge was never waited out, so extraction ran against the interstitial. Made worse by a User-Agent mismatch (headed "Chrome" vs headless "HeadlessChrome"), which invalidated cf_clearance and re-triggered the challenge.
3. Wrong stories URL — /me/stories/public now redirects to the Drafts tab; the published list is at ?tab=posts-published.
4. Login detected from cookie presence only — Medium sets uid/sid for anonymous visitors too, so the page rendered logged-out while the code thought it was authenticated.
5. Outdated selectors for the stories list.

What changed

• Re-inject a full storageState snapshot on every launch (carries session cookies through the headed→headless switch and across restarts).
• One consistent User-Agent for login + headless runs, plus an explicit Cloudflare wait after each navigation.
• Correct published-tab URL and more robust article extraction.
• Auth detected from the actual page (absence of the header Sign-in button).
• Headless background by default; the visible window opens only for the one-time login, and the other tools trigger auto-login on demand.

Bonus: fixes the documented Google-login limitation

The README notes Google login sessions couldn't be persisted. With the persistent profile + storageState re-seeding, a Google login is captured and reused silently in the background — verified end-to-end (get-my-articles returns all published stories headless, no window, no re-login).

Testing

• Reproduced the session-cookie drop with a minimal persistent-context add/close/reopen test.
• Verified end-to-end: after a one-time login, get-my-articles returns the full published list headless with no visible window and no re-login.
• npm run build / tsc --noEmit pass.

Notes

• The User-Agent pins a Chrome major version to match the bundled Chromium; it may need a bump on future Playwright upgrades.
• The saved session file contains auth cookies and should be treated as sensitive (kept in the user's home dir, git-ignored).