Add HTTP headers to all requests

[PGijsbers]

This allows us to better understand the traffic we see to our server API. It is not identifiable to a person.

Motivated after seeing large spikes in traffic recently. Some of them coming from requests/1.x.x agents - which leaves us wondering if this is from Python API users (potentially indirectly through e.g., the automlbenchmark), or scripts/bots that don't identify themselves as such.

On top of this PR, I think we might want to make our user-agent configurable so that services building on top of openml-python can identify themselves (e.g., automl benchmark, experiment bots). But I don't want to get into that right now.

I updated the unit tests that would otherwise crash. I explicitly did not add a check for the agent with these tests.
To me it feels that would be outside the scope of those tests. In general, it's a thin line the way its set up - should our tests even fail if they would just keyword argument instead of a positional one? But that too felt out of scope, so I change it to a (in my opinion) slightly better version of the same tests.

I do not feel this warrants its own unit test. If I should make one, could you recommend an approach? Just mocking the requests library and then calling _api_call for all types of requests? I just thought that wasn't particularly useful.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 83.82%. Comparing base (923b49d) to head (1a0e654).
Report is 1 commits behind head on develop.

Additional details and impacted files

@@ Coverage Diff @@## develop #1342 +/- ##
============================================
+ Coverage 32.94% 83.82% +50.87% 
============================================
Files 38 38 Lines 5254 5261 +7 ============================================
+ Hits 1731 4410 +2679 + Misses 3523 851 -2672 

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[PGijsbers]

pre-commit fail is a false negative, as the pre-commit.ci job does not fail (and neither does my local). Speaking of, we can probably remove the pre-commit/run-all-files workflow. I do not see what it accomplishes now that we have pre-commit.ci.

[LennartPurucker]

LGTM!

[LennartPurucker]

+1, let's remove pre-commit/run-all-files.