Specify User-Agent on every HTTP request (#3628)

[igrep]

Fix: #3628

Wrap up functions in Network.HTTP.Client and Network.HTTP.Simple to
add User-Agent to the request object.

• Any changes that could be relevant to users have been recorded in the ChangeLog.md
• The documentation has been updated, if necessary.

Testing the change

I confirmed some requests are fixed by stack setup, stack install, stack new.
But I haven't tested all HTTP requests made by stack.

For reviewers

I'm not sure especially with the name and place of the wrapper module (and the name of its functions).
Feel free to modify for better module structure.

[decentral1se]

This looks like a solid approach to me! Perhaps, to help avoiding using the not wrapped function, we can use a similar HLint trick like in https://github.com/commercialhaskell/stack/blob/master/.hlint.yaml#L44 - where a warning is produced (and the CI will fail) if the wrong function is used - assuming we always want these headers to go out? I assume so! Thanks for the patch!

[igrep]

@lwm Thanks for advice!

Perhaps, to help avoiding using the not wrapped function, we can use a similar HLint trick like in https://github.com/commercialhaskell/stack/blob/master/.hlint.yaml#L44

You mean, should I rename functions exported by Network.HTTP.StackClient to make them different from functions in Network.HTTP.Simple and Network.HTTP.Client?
It looks like that hlint can suggest only when lhs and rhs are literally different as their de-qualified names.
I'm new to .hlint.yaml, so tell me if I get something wrong!

If I should rename, I'm going to prefix stack to the new functions: stackHttpJSON, stackHttpLbs, etc.

[decentral1se]

I'm not exactly sure either ;) I thought you could pass full import paths (like over here) and get away with it. If you can't do that, maybe the renaming with the prefix is a better option in general - less magic!

[mgsloan]

Good stuff @igrep , thanks a bunch!

Cool idea bout using hlint to reduce likelihood of regressions, @lwm! Indeed I'm not sure if HLint can distinguish where identifiers are imported from. Seems like it'd be a major weakness if not, I'm guessing it can do this. I've opened #3700 to track implementing the idea.