Streaming output guardrails: support strict (400) and lenient (topic-change) rejection modes

[coderabbitai]

Summary

Currently, apply_stream_output_guardrails in src/proxy/handlers/format_handler.rs only supports a single hard-rejection path that returns a GatewayError::Validation error. Error handling should be extended to support two configurable modes:

1. Strict mode – return a 400 Bad Request status code when the guardrail blocks streaming output.
2. Lenient mode – instead of an error, return a simulated "Let's change the subject" (topic-deflection) response, giving a softer guardrail experience.

Background

Raised during review of PR #108 (comment: #108 (comment)) as a follow-up to the streaming output guardrail implementation.

Requested by

@bzp2010

[coderabbitai]

🔗 Related PRs

#23 - feat(provider): add stream bridged and native implementations [merged]
#25 - feat(provider): add llm sdk entrypoint for streaming [merged]
#27 - feat(provider): add anthropic chat format for streaming [merged]
#92 - feat: unified chat handler [merged]
#104 - feat(guardrail): integrate into chat handler [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!