v4.4.0

Changed

• One error format for every tool: <Title>: <what happened>. <what to do> Details: <details>, e.g. Zone not found: ... or Rate limited: .... Titles replace the old codes (invalid_input, rate_limited,
  ...). Details carry the API's message, the HTTP status, or the error
  text, on every error.
• Search errors name the query, fetch errors name the URL.
• A missing zone, a usage limit and a failed sign-in are reported as
  such, never as a bad input.
• scrape_batch returns one error per URL even when every URL fails. A
  batch that fails as a whole (job failed, timed out, rejected) is one
  error.
• search_engine_batch returns one error when every query fails the same
  way (e.g. a bad token), instead of a copy per query.

Fixed

• scrape_as_html and extract report a failed page (unknown domain,
  site error, empty page) as an error instead of empty content.
• An unknown domain is reported as Site not found on fetch tools too.
• search_dataset with an unsupported filter says Filter rejected
  instead of reporting a service outage.
• Browser tools with a bad token report Sign-in failed.
• Bright Data's own error messages are no longer marked as untrusted web
  content; only site error pages from scrape_as_html and extract are.
• A page in a misread character encoding (e.g. Shift_JIS) is returned
  with a note instead of being called a PDF or binary file.
• URLs with a Unicode domain (e.g. https://münchen.de/) work: the host
  is sent as punycode, results keep the URL as given.