feat: add runtime extra HTTP headers support

[nazq]

Summary

• Adds extra_headers() builder method to LLMBuilder for runtime-configurable HTTP headers
• Enables use cases like Cloudflare Access authentication tokens for secure vLLM endpoints
• Headers are applied to all HTTP requests made by OpenAI-compatible providers

Note: This PR is based on #94 (streaming with tools support) because the new chat_stream_with_tools() methods also make HTTP requests that need extra headers support.

Use Case

use std::collections::HashMap;
use llm::builder::LLMBuilder;

let headers = HashMap::from([
    ("CF-Access-Client-Id".to_string(), "my-client-id".to_string()),
    ("CF-Access-Client-Secret".to_string(), "my-secret".to_string()),
]);

let llm = LLMBuilder::new()
    .backend(LLMBackend::OpenAI)
    .base_url("https://vllm.example.com/v1/")
    .api_key("not-needed")
    .model("meta-llama/Llama-3.1-8B-Instruct")
    .extra_headers(headers)
    .build()?;

Changes

• src/builder.rs - Added extra_headers field and builder method
• src/providers/openai_compatible.rs - Store and apply extra headers in requests
• src/backends/openai.rs - Pass extra headers to provider
• Other backends updated to pass None for compatibility

Test plan

• Build passes
• Clippy passes
• Unit tests for extra_headers functionality added
• Manual testing with Cloudflare-protected endpoint

[Copilot]

Pull request overview

This PR adds runtime-configurable HTTP headers support for LLM providers, enabling use cases like Cloudflare Access authentication. The PR also includes streaming with tools functionality from PR #94.

Key Changes:

• Added extra_headers() builder method to configure runtime HTTP headers
• Implemented chat_stream_with_tools() for streaming responses with tool call support
• Added StreamChunk enum for unified streaming events (text, tool calls, completion)

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/builder.rs	Added extra_headers field and builder method; passed to OpenAI and Cohere backends
src/providers/openai_compatible.rs	Stored extra_headers field; applied headers to all chat/streaming requests; implemented chat_stream_with_tools() with SSE parsing
src/backends/openai.rs	Passed extra_headers to provider; added extra_headers to hosted tools endpoint; delegated chat_stream_with_tools() to provider
src/backends/mistral.rs	Added None for extra_headers parameter (not exposed)
src/backends/groq.rs	Added None for extra_headers parameter (not exposed)
src/backends/cohere.rs	Added None for extra_headers parameter (not exposed)
src/backends/huggingface.rs	Added None for extra_headers parameter (not exposed)
src/backends/openrouter.rs	Added None for extra_headers parameter (not exposed)
src/backends/anthropic.rs	Implemented chat_stream_with_tools() with Anthropic-specific SSE parsing; handles empty tool arguments correctly
src/chat/mod.rs	Added StreamChunk enum and chat_stream_with_tools() trait method
src/resilient_llm.rs	Added resilient wrappers for new streaming methods
tests/test_backends.rs	Added comprehensive integration tests for streaming with tools (Anthropic backend)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[graniet]

Hey @nazq, thanks for your work. Could you resolve the conflict?

[nazq]

Sure. Give me a couple hours

[nazq]

Hey @nazq, thanks for your work. Could you resolve the conflict?

Done