fix: services/httpd: parse correctly Accept header with extra test cases

[jsign]

Closes #14296

Until the standard lib decides to include a standard function to correctly parse Accept headers, we can rely on goautoneg library which is a Github port used by Prometheus.

I also added sub-tests as Table Tests for multiples scenarios around this problem.

Adding a new dependency for this amount of code may hurt a bit, so maybe we could add the code directly too.

• CHANGELOG.md updated with a link to the PR (not the Issue)
• Rebased/mergeable
• Tests pass
• Signed CLA (if not already signed)

[jsign]

The failed test case seems to be a reported flaky test

[jsign]

@russorat , I rebased it to 1.8 to keep it clean.
Is 1.8 OK for merging or do you prefer on master?

[jsternberg]

Can you copy the code from the dependency into the code as an unexported function? Make a reference that the code is based on that person's original library and a link to the license as ours would become a derivative work so we would need to give credit as part of the license.

There's things I want to change in both your code and the dependency code to ensure this is done correctly and won't require fixes in the future and it is easier to do that if we're not pulling in a dependency we don't control for something so small.

[jsign]

@jsternberg, pushed changes in the direction you mentioned.

[jsign]

@jsternberg, I applied your suggestions, and also switched to performance mode since the code had many allocations that could be avoided. And considering this is a hot-path, deserved some extra attention.

Some points about the changes (apart from your comments):

• make csvFormatter stateless
• all known content types and formatters are defined as vars, so to a ton of allocations. This is the reason why I made csvFormatter stateless (since it wasn't).
• avoided string concats to save extra allocations (that's why full exist apart from splited parts)
• added * test case
• in case of */* the switch isn't biased to a definite option. This is strictly correct (since any option complies to request), but maybe another default is wanted.
• as can be seen, adding future content types is much less verbose
• rebased again to 1.8

[jsternberg]

Sorry I haven't had time to review this again. I will review by end of day tomorrow.

[dgnorton]

@jsternberg can you review this for the 1.8 release coming up?

[jsternberg]

I reverted the changes to the formatters and modified the creation of those formatters to a factory method. I think that the state is actually used by the code so a global instance doesn't work, but I also don't remember well enough to know for certain so I don't want those to change.

I am going to wait for the tests and run some myself and then merge this.