Investigate header percent-encoding with clarifications in recent OpenAPI versions

[czechboy0]

The question of percent-encoding header values over the evolution of the OpenAPI specification has been a bit messy, and 3.1.2 adds to it again: https://spec.openapis.org/oas/v3.1.2.html#appendix-d-serializing-headers-and-cookies

This issue tracks tracking down the correct behavior, and possibly defaulting to different behavior based on the version of the OpenAPI document - to avoid making backwards-incompatible changes.

This might require extending OpenAPI Runtime to allow customizing whether to percent-encode header values or not (maybe can even be added to the Configuration struct), and try to use the correct behavior on the newer version of OpenAPI docs where this behavior was clarified.

[czechboy0]

Seems 3.0.4 and 3.1.1 changed that, but we still percent-encode. We should stop doing that, but the question is how to do it correctly without potentially breaking existing adopters.

We could add an opt-in flag, or an opt-out flag (and fix the behavior by default), we could switch the behavior based on 3.0.3 vs 3.0.4 and 3.1.0 vs 3.1.1. Note the new ones (3.1.2, 3.2.0) even more clearly state to not perform percent encoding, and for header values that include non-standard characters, to perform encoding (eg base64) at the application level.