fix: consistent proxy auth handling

[barjin]

Extracts and reuses the getBasic helper function.

Uses this helper to generate the Basic authentication header in the proxy hooks (before, this was done further down the line by http2-wrapper, see the related issue).

[azerum]

Thanks for fixing! This has robbed me a few hours with proxy provider that used ; and . for passing params

There might be a potential issue: the code uses URL to parse url string and decodeURIComponent to retrieve the original username/password from parsed URL

URL uses percent-encoding scheme different from that of decodeURIComponent. The only difference I've found so far is using % in username. It is allowed, but decodeURIComponent breaks with it

new URL('http://us%r:[email protected]').username
'us%r'
decodeURIComponent('us%r')
Uncaught URIError: URI malformed

Perhaps most providers don't use %, but technically we would use inverse of percent-encoding URL does - percentDecodeBytes() from whatwg-url. Not sure if dependency is worth it

P.S.

For URL spec, see https://url.spec.whatwg.org/#concept-basic-url-parser, section authority state. Here's the step-by-step parsing of URL:

1. https://url.spec.whatwg.org/#api-url-parser

2. (Step 3) https://url.spec.whatwg.org/#concept-basic-url-parser

3. The last link as a state machine of parser. Not super readable, but we can see how username/password is parsed in "authority state" section

4. In "authority state", here's the loop that parses username/password character-by-character:

4. For each codePoint in buffer:
   4.1. If codePoint is U+003A (:) and passwordTokenSeen is false, then set passwordTokenSeen to true and continue.
   4.2. Let encodedCodePoints be the result of running UTF-8 percent-encode codePoint using the userinfo percent-encode set.
   4.3. If passwordTokenSeen is true, then append encodedCodePoints to url’s password.
   4.4. Otherwise, append encodedCodePoints to url’s username.

Note that username/password user special set of characters for percent-decoding, which does not include %, that's why encoding does not touch it