Add charset_normalizer detection.

[lovelydinosaur]

Use charset_normalizer to auto-detect character encodings in cases where no Content-Type: text/...; charset=... is included.

See #1657 and https://github.com/tomchristie/top-1000 for some evidence-led rationale behind this change.

[Ousret]

There are cases where the detection works just fine with small content. I would suggest silent the warning instead.

(target: x.encode("utf_8"))
* Using the following `Qu'est ce que une étoile?`
chardet detect ISO-8859-1
cchardet detect IBM852
charset-normalizer detect utf-8

  * Using the following `Qu’est ce que une étoile?`
chardet detect utf-8
cchardet detect UTF-8
charset-normalizer detect utf-8

  * Using the following `<?xml ?><c>Financiën</c>`
chardet detect ISO-8859-1
cchardet detect ISO-8859-13
charset-normalizer detect utf-8

  * Using the following `(° ͜ʖ °), creepy face, smiley 😀`
chardet detect Windows-1254
cchardet detect UTF-8
charset-normalizer detect utf-8

  * Using the following `["Financiën", "La France"]`
chardet detect utf-8
cchardet detect ISO-8859-13
charset-normalizer detect utf-8

WDYT?

[lovelydinosaur]

I went with that originally, and a couple of the tests with small amounts of content returned results I wasn't expecting. If apparent_encoding is None, then we'll end up decoding it with 'utf-8', errors='replace', whichI figure is a pretty reasonable default for the corner case.

[Ousret]

Alrighty, that is reasonable. 👍
In another matter, you may run the detection anyway and check if the result has a SIG/BOM, that could be reasonable too. And discard it if len(content) < 32 and best_guess.bom is False.

results = from_bytes(content)
best_guess = results.best()

if best_guess.bom:
    ...