fix(response-ratelimiting): fix missing usage headers for upstream

[t-yuki]

Summary

response-ratelimiting plugin should send usage headers to upstream server.
But in Kong 3.8, there are no usage headers such as X-RateLimit-Remaining-Videos: 10 for upstream requests.

In this change, it coming back usage headers for upstream.

Checklist

• The Pull Request has tests
  • It seems to be already there are tests for this feature but are there passing? https://github.com/Kong/kong/blob/master/spec/03-plugins/24-response-rate-limiting/04-access_spec.lua#L505
• A changelog file has been created under changelog/unreleased/kong or skip-changelog label added on PR if changelog is unnecessary. README.md
• There is a user-facing docs PR against https://github.com/Kong/docs.konghq.com - PUT DOCS PR HERE

Issue reference

Fix #13682

[ADD-SP]

@t-yuki Thanks for your contribution! Would you mind adding a regression test on this issue? I think we have missed this test for a long time.

And also a changelog entry.

[ADD-SP]

Since this is a new issue introduced in 3.8 and should be a small fix, I added this to the 3.9 milestone.

Internal ticket: KAG-5447

[ADD-SP]

@t-yuki Could you fix failure tests?

[t-yuki]

@t-yuki Could you fix failure tests?

OK, I'll try to continue within a week.

[gszr]

Since this is touching a recent nontrivial refactor, we should carefully verify it doesn't revert important bits of said refactor, fixing one but introducing other issues. cc @ADD-SP

[ProBrian]

I see that wait time granularity change from milli seconds to dozens of seconds, is there a reason? I think this will make the CI test cases spend much more time.

[t-yuki]

Overall, changing millis to seconds was due to avoid unstability of tests.

1. As far as I tried, the nginx process and the test process are not synced well in the ms order. Thus, when the test process sleeps with ngx.sleep(1-ms) to on-set ss.000 ms, the nginx process might be still at ss.998 ms.
   • Adding several jitters may relax this problem, but in my environment, it requires several hundred ms (WSL2+Docker)
2. Some of test takes several hundred ms to run, especially, iteration based-tests such as

   kong/spec/03-plugins/24-response-rate-limiting/04-access_spec.lua

   Line 483 in 4c3b9cd

   for i = n+1, ITERATIONS do

   must be completed 10 requests in 1secs, that is not stable.

In this change, wait(reset) function will only sleep if current time is at 50-59s or counter-reset is required. Since most of tests are separated into different limit counters, the CI will sleep only 1 to 3 times during tests.