[sergeym]

Closes bddkit/bddkit#96.

• Steps 32–33: I dump the DOM (WebDriver GET /source) and I dump the DOM of "<selector>" (outerHTML via execute). Neither waits; a missing element is fatal.
• on_failure accepts dom (off by default); a failure carries a DOM diagnostic with the file path.
• Screenshots and DOM dumps are named screenshot-<n>.png / dom-<n>.html, n counting from 1 per scenario.
• e2e: a missing bddkit binary or browser now fails the suite when CI is set instead of skipping it, so the examples cannot go green unrun.

Things to know:

• Dumps are written raw. Neither the plugin nor the host masks anything (console and network dumps already behave the same), so hidden fields and tokens are in the file. The README says so. This is why dom is opt-in.
• Shadow roots and iframe contents are not in either dump.
• The host gives every step its own artifacts_dir, so the -<n> suffix does not prevent overwrites; it records the order within a scenario.

Verified: unit and stub suites, clippy, fmt; e2e (examples on Chrome and managed Firefox) against a live Chrome.