v1.0.1

Fixed

• A response is decoded the way it declares itself. PageSoup.create read the body
  of a Response as UTF-8 with errors="ignore" and never looked at what the page said
  its charset was, so every multi-byte character on a non-UTF-8 site was silently
  dropped — a GBK page returned an empty title rather than a wrong one. The charset now
  comes from the first source that declares a usable one: an explicit encoding=
  argument, then the response's Content-Type header, then a <meta charset> near the
  top of the markup, then UTF-8. A charset the server names but Python cannot load falls
  through to the next candidate instead of raising.

  Deliberately not response.encoding: requests fills that with ISO-8859-1 for any
  text/* response that declared no charset, so preferring it would mojibake exactly the
  pages this fixes. Scraper._peek does read it, which is why diagnosis and the parsed
  soup could disagree about the same bytes.

• parser survives a Response. PageSoup.create recursed into its own bytes
  branch without passing parser on, so ScraperConfig.parser and Scraper(parser=…)
  had no effect on get_soup, post_soup or make_soup(response) — the only paths a
  caller uses — and everything was parsed with lxml.

• A challenge is no longer written to a download target. stream_to blanked the body
  before diagnosis ran, so a challenge interstitial — which arrives with a 200 and a body
  — was streamed to the caller's path and accepted. get_file produced a file that was
  really a Cloudflare page, indistinguishable from the asset once the response was gone.
  The opening bytes are now held back and diagnosed before the file is created, and the
  abort signal is honoured while they are buffered as well as while the rest is written.