test(e2e): re-enable 5 skipped browser tests, add coverage [follow-up to #283]

[EmersonBraun]

Summary

Removes the `test.describe.skip` wrappers in the React and multi-agent E2E specs, expands assertion coverage, and proves the browser side end-to-end.

Local result: 10 passed, 0 skipped, 0 failed. Up from 4 passed + 5 skipped on PR #282.

Why this was blocked

PR #282 introduced Playwright coverage for all four example apps. Two of the four (react, multi-agent) had their browser-side tests skipped because `@agentskit/core` imported `node:fs/promises` at module load — the React root never mounted, so there was no UI to assert against.

PR #283 fixed that core issue by relocating Node-only helpers to `@agentskit/memory` and `@agentskit/cli`. This branch is built on top of #283 and removes the skips.

What changed

Selectors

Switched from `getByPlaceholder(/type|message/i)` to `locator('[data-ak-input]')`. The example apps use custom placeholders ('Ask about the weather in any city...', 'Give the planner a task to delegate...') that the regex didn't match. The `data-ak-input` attribute is the headless contract — stable across placeholder changes.

Coverage expanded

• example-react: 5 tests (server / mount / typed input reflected / submit shows user message / first turn triggers tool call)
• example-multi-agent: 3 tests (server / mount / submit)
• example-ink: 1 test (unchanged)
• example-runtime: 1 test (unchanged)

Dependencies

Order	PR
1st	#283 (core browser-compat fix) — merge first
2nd	#282 (E2E infrastructure) — already provides playwright config + helpers + workflow
3rd	This PR (#286) — re-enables tests + better selectors

This branch already includes the fix from #283 (merged in), so it's self-contained — but if #283 lands first on `main`, this PR's diff shrinks naturally.

Verification

```bash
pnpm --filter "./packages/*" build
pnpm exec playwright test
```

Local run output (10/10 passing, 17.2s):

```
✓ ink — renders the chat container
✓ multi-agent — dev server serves index HTML
✓ multi-agent — chat UI mounts
✓ multi-agent — submitting a prompt produces output
✓ react — dev server serves index HTML
✓ react — chat UI mounts
✓ react — typed input is reflected in the textarea
✓ react — submitting renders the user message
✓ react — first turn triggers a tool call
✓ runtime — executes and produces assistant output
```

Test plan

• All previously-skipped tests now run
• No `test.skip` left in the e2e directory
• Selectors use stable headless attributes (`data-ak-*`)
• Demo adapters (deterministic) make assertions reproducible
• CI confirms the 10-test pass after merge order is respected

Refs #251 #281 #283 #211