[Bug] Console Go gateway rejects OpenAI-compatible client metadata with `extra inputs are not permitted` (strict schema, extra="forbid")

Description

Describe the bug

The OpenCode "Console Go" API gateway (https://opencode.ai/zen/go/v1) validates the messages array with a strict schema (pydantic extra="forbid" behavior). Any non-standard field attached to a message is rejected with HTTP 400:

[invalid_request_error] Extra inputs are not permitted, field: 'messages[1].agent', value: 'cli'

This breaks interoperability with OpenAI-compatible clients that attach extra metadata to messages (e.g. agent, reasoning, messageId, model, traceId, usage, extra_fields, etc.). Standard OpenAI-compatible behavior is to ignore unknown fields, not reject them.

To Reproduce

Send a minimal chat completion, then the same request with one extra field:

# 1) Clean payload -> HTTP 200 (model, key, format all fine)
curl https://opencode.ai/zen/go/v1/chat/completions \
  -H "Authorization: Bearer $OPENCODE_KEY" -H "Content-Type: application/json" \
  -d '{"model":"glm-5.2","messages":[{"role":"user","content":"hi"}],"max_tokens":16,"stream":false}'

# 2) Same request + one extra field on a message -> HTTP 400
curl https://opencode.ai/zen/go/v1/chat/completions \
  -H "Authorization: Bearer $OPENCODE_KEY" -H "Content-Type: application/json" \
  -d '{"model":"glm-5.2","messages":[{"role":"user","content":"hi"},{"role":"assistant","content":"hello","agent":"cli"}],"max_tokens":16,"stream":false}'
# -> 400 {"error":{"message":"Error from provider (Console Go): ... Extra inputs are not permitted, field: 'messages[1].agent', value: 'cli'"}}

Expected behavior

OpenAI-compatible endpoints should tolerate and ignore unknown message fields, so clients that attach metadata (agent ids, tracing, reasoning flags) keep working. The request in step 2 should return 200 like step 1.

Actual behavior

HTTP 400 with Extra inputs are not permitted for any non-standard field on any message.

Environment

• Gateway: https://opencode.ai/zen/go/v1 (Console Go)
• Model: glm-5.2
• Client: WorkBuddy App 5.3.13 (sends requests via its backend, which attaches internal metadata fields)
• Observed on: macOS

Suggested fix

Relax the messages / message schema at the Console Go gateway to extra="allow" (ignore unknown fields), matching standard OpenAI-compatible server behavior. Optionally document which fields are required vs optional.

Additional context

This is not specific to one client — any OpenAI-compatible proxy/agent that forwards internal metadata will hit the same 400. A clean payload works perfectly (step 1 returns valid reasoning_content too), confirming auth + model + base format are correct; only the strict schema blocks it.

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

Start by running the two curl requests against the Console Go gateway at https://opencode.ai/zen/go/v1 and determine whether its message schema is implemented in this repository. Trace the gateway's message validation and add coverage for an unknown message field; done means the metadata-bearing request is accepted without the existing HTTP 400 error.