Remove donor-repo leftovers that described other sites as this one

[jehrr]

Leftovers from the sibling repos this one was bootstrapped from (a blogging site, a Q&A site, a rental site), in text that read as a description of polymarket.com.

Fixed

• User-visible logs: Selenium said "No property cards appeared … this search", pyppeteer "No answer cards appeared … this feed". Both now say "No event tiles … this listing", which is what Playwright already said.
• pyppeteer block advice said its bundled Chromium was "the build this site was measured refusing". The README and the engine docstring measure it served here. It now names build 117.0.5938.0 and still points at --chromium-path.
• Sidecar stop_reason: an unrecognised refusal was blocked_bot-or-not, a sibling's challenge-page name. It is now blocked_not-served, matching the log line "The site did not serve this request". Nothing in the repo branches on the old value.
• scraper_api_client.py: the block message cited a sibling's 2026-09-10 datacentre refusal, which contradicts this module's own 2026-09-16 measurement. The --help description mentioned a sibling's archive payload and author pages. The --url help ended in a broken sentence about "a topic's answers".
• output_writer: the mode default in run_meta/finish_run was "topic", which is not in UNIQUE_BY_SKU_MODES, so diff_runs.py would have refused it. The engines always pass a mode, so no run was affected. The default is now "markets".
• diff_runs.py: the --price-tolerance-pct help (clap counts, text_chars) and the host-mismatch message ("serves one story from several addresses") now describe markets. So do the comments about per-story language, the missing lifecycle bucket and the source_changed payload.
• captcha_solver.py: removed a paragraph saying this site refuses headless browsers with a 394-byte "Access Denied" page. It now says what the README measured: headless and headful identical, no challenge met, and Cloudflare/Turnstile as the plausible block.
• smoke_test.py: in the docstring, the claps bullet, the sign-in-widget bullet and a bullet about a test this repo lacks are replaced with this repo's facts. The docstring also claimed the fixtures pseudonymise author handles, but make_fixtures.py rewrites nothing. The worker-pool test fed medium.com/tag/.../archive URLs and counted "days". It now feeds event-page URLs and asserts the same things. Two comments told a sibling's live-run story as this repo's.
• .dockerignore: vrbo_products.* is replaced with this repo's --out default (polymarket_markets.*, polymarket_markets_*.*, as in .gitignore).
• Smaller fixes: the bug-report placeholder quoted a sibling's "96 products", SECURITY.md said the project has no releases (v0.1.0 and v0.1.1 exist), and a tests.yml comment said "feed".
• CHANGELOG: added a Fixed entry under the existing [Unreleased]. Released sections are untouched.

Kept, deliberately

• The guard list of sibling names and vocabulary in test_no_file_describes_another_site, including its diff_runs.py/CHANGELOG allowlist. It exists on purpose.
• diff_runs.py's post-mortem comment naming the columns it used to track. It is explicit provenance, allowlisted by the suite.
• The captcha_solver notes that name a sibling as a sibling (the image-captcha solver not ported, and the farfetch reCAPTCHA-modal note), and the .gitignore craigslist note.
• The released CHANGELOG sections.
• The runs without an account badge. The README measures every path with no account, key or proxy, and the badge order already matches the family order.
• Repo description, topics and homepage were checked and have no donor leftovers.

Suite: smoke_test: OK before and after, with Playwright installed so the edited worker-pool group ran. ci_checks.py --secret-check and --help-check --sample-check are green.

🤖 Generated with Claude Code