fix(docker): include Playwright headless shell

[reallav0]

Summary

• copy Playwright's chromium_headless_shell-* cache into the runtime user's cache
• add a Dockerfile regression check for both Chromium artifacts

Root cause

The image runs as appuser, but only copied the full Chromium cache from root. Headless crawls resolve chromium_headless_shell-*, so the Docker server exited at startup when that directory was missing.

Fixes #2064.

Validation

• git diff --check
• standard-library Dockerfile regression check passed

Not run: Docker image build (Docker daemon unavailable); focused pytest could not be installed because the full project dependency set exhausted available disk space.

[SohamKukreti]

Thanks for the quick fix! I dug into #2064 independently and can confirm both the root cause and that this fix works — verified against the published images.

Root cause (confirmed):

• playwright install --with-deps installs two Chromium artifacts in root's cache: chromium-1228 and chromium_headless_shell-1228. The old copy glob chromium-* (hyphen) never matched the underscore-named headless shell, so it's missing from appuser's cache in both the 0.9.0 and 0.9.1 images.
• This was harmless until 0.9.1: commit c34ffc0 (#2051) stopped passing channel="chromium" for the default config, and with no channel, Playwright ≥1.49 launches headless mode with the headless shell — the one binary the runtime user never had. Boot crash → restart loop.

Fix verification: I injected chromium_headless_shell-1228 into /home/appuser/.cache/ms-playwright/ in a stock unclecode/crawl4ai:0.9.1 container (byte-for-byte what this PR's copy produces) and the container boots cleanly — Application startup complete, /health returns OK, and the permanent browser pool runs with a chrome-headless-shell process. Image size cost is +262 MB uncompressed (~7% on a ~3.9 GB image).

Two things this PR gets right that are worth preserving in review:

• Keeping both binaries (not just the headless shell): full chromium-* is still required for headless=false requests — the image ships Xvfb precisely for headed mode. Copying only the shell would fix the boot crash but break headed crawls at request time.
• Keeping cp rather than mv: crawl4ai-doctor runs as root after this line and needs root's copy intact.

One non-blocking nit on the test regex (inline). LGTM otherwise, happy to see this un-drafted and merged.