fix: Impersonating JWT (obo) header not propagated correctly for OpenAI requests.

[mgiannakopoulos]

WIth this change we:

• Revert the ability for OpenAI clients to dynamically pass api_token and obo_token values (we can achieve this by providing a custom http_client in the factory methods and ensuring that we provide the bearer auth token as another header compared to the default Authorization header [change will happen in a subsequent PR and once Unified Model Gateway accounts for this fact as well]).
• http_client parameter won't be in a deprecated state anymore. We will continue using it as expected.
• Removal of timeout parameter, we get the values for timeouts from the http_client parameter.
• Initialization for OpenAI clients follows the old pattern not breaking the Impersonating JWT (OBO token). The current implementation is not working because OpenAI re-writes all the header values when we try to modify the Authorization header as well.

[nunogoncalves03]

Does this PR affect chat factory initialization time for analyst servers?

[mgiannakopoulos]

Does this PR affect chat factory initialization time for analyst servers?

No, it doesn't. The existing gains should be present and intact with those changes.