Set blank headers so they are not automatically resolved by go-gh

[samcoe]

This PR fixes two bugs dealing with the handling of headers for http requests.

1. Requests using an api.Client will now properly send the "User-Agent" header instead of having it overwritten by go-gh.
2. Requests from the api command can pass an alternative Authorization header.

[csainty]

I've got a problem today where I can no longer pass an alternative Authorization header to gh api calls. It is ignored and my logged in user is used instead.

Am I right in thinking this PR would resolve that?

[mislav]

Confirmed that this fixes things for me; thank you! In the future, I propose that we make this more declarative; i.e. that we don't have to explicitly set headers to blank strings just so that they can be set to a correct value later. But I guess that might require a change to go-gh behavior?

[mislav]

The semantics for passing a blank string to get the desired (non-blank) result in the end is confusing to me; would there be a way to make this clearer as follow-up?

[samcoe]

@mislav I agree it is confusing. I have a few ideas on what we could do in go-gh to make this better. I think making the changes here are be the quickest solution to fix the bug and once the bug is fixed we can take a more thorough look at the options we have to make it less confusing.