Upgrade httpx

[dalf]

What does this PR do?

Upgrade to httpx 0.21.2

Change in searx.network.client:

• remove close_connections_for_url : httpcore closes the connections now.
• keep AsyncProxyTransportFixed to map the exception from httpx_socks to httpx exception WITH the "request" object.
• remove AsyncHTTPTransportFixed: the retry mecanism has moved to searx.network.network to decrease the number of access to the private API of httpx / httpcore.

httpx.RemoteProtocolError happens when a server closes an HTTP/2 connection. In the master branch, AsyncHTTPTransportFixed closes all connections. In this PR, the httpx.AsyncClient is closed. It makes no difference except when different proxies are specified, for example:

proxies:
  https://bing.com/: socks5h://127.0.0.1:1337
  all://: socks5h://127.0.0.1:1338

With this configuration, and with an httpx.RemoteProtocolError exception coming from the proxy on the port 1337:

• in the master branch: connections using the proxy on the port 1337 are closed. Connections using the proxy on the port 1338 are untouched.
• in this PR: connections using the proxy on ports 1337 and 1338 are closed.

@unixfox @tiekoetter @paulgio I don't know you use this configuration?

An alternative configuration not impacted by this PR:

outgoing:
	proxies:
		all://: socks://127.0.0.1:1337	

engines:
  - name: bing
    engine: bing
    shortcut: bi
	proxies:
		all://: socks://127.0.0.1:1338

---

This is WIP, todo list:

• use a released httpcore version
• check for memory leak over a week
• check for error log over a week:
  • in the master branch, there is a close connections + retry mechanism in a case of specific error.
    this a workaround to fix some of the httpcore issues
  • the latest httpcore commit seems to have fix most of them
  • the change in this commit remove the "close connections + retry"
  • an error log over a long time can confirm that httpcore has been fixed

Why is this change important?

• Remove a bunch of hack to fix an old httpcore version
• With this change, httpx can be upgraded easily (= just change requirements.txt)

How to test this PR locally?

A lot of changes have been made to httpx, httpcore and anyio since the last upgrade, a full check is require over a long time.

what can be done locally:

• check the error for various HTTP error (domain not found)
• check that HTTP proxy works
• check that SOCKS proxy works
• check that HTTP2 connection works
• check this scenario:
  • make a HTTP2 request
  • wait for the server to disconnect
  • maka another HTTP2 request

Author's checklist

Related issues

Related to #589

[mrpaulblack]

I tested this on my instance over a week:

Memory Increased over the span of 3 days by 13megabytes which is IMO margin of error.

Also with this PR memory usage dropped a little and the performance is the same or even slightly better in general IMO

[dalf]

The only missing part is a new release of httpcore, after that the PR is ready to merge.

[unixfox]

I don't use proxies anymore, but I do I've plans to add some in the future but not for the default engines, for some engines that block public instances like qwant.

[mrpaulblack]

@unixfox I have actually qwant on my instance enabled, because they never IP block it tbh (and on top of this qwant is really fast; around 400ms per request)...

[HLFH]

Besides the missing httpcore release, the branches upgrade-httpx and/or test-uwsgi are ready to merge.