bug(proxy): valid HTTP/1.1 h2c upgrade requests from JetBrains/Ktor are rejected with 400

[NormanGeyerProdatis]

Pre-flight checklist

• I searched existing issues and confirmed this is not a duplicate.
• I am running a supported codex-lb version (see Releases).
• I can reproduce the problem with the steps below.

codex-lb version

1.23.0

Deployment method

Docker (ghcr.io/Soju06/codex-lb)

Client used against codex-lb

Other (describe below)

ChatGPT account plan(s) involved

Business

Model(s) involved

gpt-5.6-luna

What happened?

A direct OpenAI-compatible Responses API request from JetBrains/Ktor fails before request authentication or payload processing.

The client sends an HTTP/1.1 request with the standard cleartext HTTP/2 upgrade headers:

Connection: Upgrade, HTTP2-Settings
Upgrade: h2c
HTTP2-Settings: <base64url settings payload>
codex-lb responds:
HTTP/1.1 400 Bad Request
Invalid HTTP request received.
The same JSON request is otherwise valid and contains a normal Responses API payload, including model, input, and stream: true.

What did you expect to happen?

A server which does not elect to perform the h2c upgrade should continue to process the request as HTTP/1.1, or otherwise return a protocol-appropriate upgrade response. The presence of valid HTTP/1.1 upgrade headers should not cause the whole request to be rejected as an invalid HTTP request.
In particular, the OpenAI-compatible /v1/responses endpoint should remain usable by HTTP clients that opportunistically offer h2c.

Steps to reproduce

Run codex-lb 1.23.0 in Docker on port 2455.
Configure IntelliJ AI Assistant with an OpenAI-compatible provider: http://127.0.0.1:2455/v1
Select OpenAIAPI/gpt-5.6-luna for Core Features.
Trigger AI commit-message generation.
Observe POST /v1/responses returning HTTP 400.

Relevant logs

Configuration / environment

Additional context

No response