brotli & zstandard support

[Do1e]

I strongly not recommend to continue using the current decoding strategy for Brotli & Zstandard.

According to my attempts, when the response compression method is br, response.text is displayed as garbled. At first I thought it was a problem of text encoding, which wasted a lot of time.

According to httpx/_decoders.py, I found that there are errors when the decoder initializes. But because of pop, nothing is done.

[tomchristie]

I think I know what's going on there, tho could you demonstrate an example against a public URL to help us talk this through?

[Do1e]

An example on Archlinux:

> conda create -n httpx python=3.12
> conda activate httpx
> pip install "httpx[http2]"
> python test.py

# test.py
import httpx

url = 'https://ip.011102.xyz/'
response = httpx.get(url, headers={'accept-encoding': 'zstd'})
print(response.headers.get('content-encoding'))
print(response.text)

If change headers={'accept-encoding': 'zstd'} to headers={'accept-encoding': 'gzip'} or install zstandard, the result will be correct.

If delete following lines in _decoders.py

if brotli is None:
    SUPPORTED_DECODERS.pop("br")  # pragma: no cover
if zstd is None:
    SUPPORTED_DECODERS.pop("zstd")  # pragma: no cover

It will report the error as expected.

[Do1e]

I realized the above porblems after trying different text encoding and carefully reading the document, which wasted a lot of time. I think it's better to give the error at first.