routing/http/client: avoid race by not using global http.Client

[hacdias]

Fixes the race condition observed in the tests in ipfs/rainbow#87.

It could've been fixed in rainbow, but that doesn't fix the source issue: we use the same global default HTTP client, while giving an option that can override its user agent. This option changes the user agent of the global variable, which may be used by multiple clients.

In addition, I updated the tests and documentations to reflect how WithUserAgent and WithHTTPClient work. The tests were passing until now by mistake and only because we were using the global defaultHTTPClient .

[codecov]

Codecov Report

Attention: 8 lines in your changes are missing coverage. Please review.

Comparison is base (79cb4e2) 65.56% compared to head (eeef41d) 65.60%.

@@            Coverage Diff             @@
##             main     #579      +/-   ##
==========================================
+ Coverage   65.56%   65.60%   +0.04%     
==========================================
  Files         207      207              
  Lines       25597    25612      +15     
==========================================
+ Hits        16782    16804      +22     
+ Misses       7336     7331       -5     
+ Partials     1479     1477       -2     

Files	Coverage Δ
routing/http/client/client.go	73.00% <69.23%> (-1.60%)	⬇️

... and 10 files with indirect coverage changes

[hacdias]

@aschmahmann I updated the options to return an error, and simplified the documentation of the options.

[aschmahmann]

Code changes look good, left some documentation suggestions/clarifications