feat(llm): add openrouter backend support with configurable API and model options

[the-cybersapien]

This pull request adds support for OpenRouter as a new LLM backend in the graphify/llm.py configuration. The main change is the introduction of a configuration block for OpenRouter, allowing users to route requests through OpenRouter's OpenAI-compatible API proxy and select from a variety of models.

New LLM backend support:

• Added an openrouter configuration to the LLM backends, setting up default parameters, environment variable keys, and documentation for using OpenRouter's API proxy with a default model of qwen/qwen3.6-flash.…options

[the-cybersapien]

Hello @safishamsi , I previously raised a PR #723 for this. We have several other backends supported now. Openrouter provides a flexible option for this use-case.
I am currently using custom env vars with ollama backend, this would simplify it a lot.

[safishamsi]

Thanks for this! OpenRouter backend is a great addition. A few things to fix before we can merge:

1. Wrong default model: qwen/qwen3.6-flash doesn't appear to be a valid OpenRouter model ID. The correct Qwen3 IDs on OpenRouter are qwen/qwen3-8b, qwen/qwen3-30b-a3b, qwen/qwen3-235b-a22b, etc. Could you double-check the default against https://openrouter.ai/models and pick one that actually resolves?

2. Zero cost: The 0.0 cost for free-tier models is correct behavior, but it should be documented in the docstring/comment so it doesn't look like a bug to future readers. Something like: # Free-tier models return 0.0 cost — this is correct.

3. Misleading comment: There's a comment that implies something about the request structure that doesn't match what the code actually does - worth cleaning up for clarity.

Once those are addressed this is ready to go!

[the-cybersapien]

Hey @safishamsi ,

I've made some fixes. Responses point by point:

1. I am actively using qwen/qwen3.6-flash It's available here: Openrouter link.
2. Updated cost to the default (qwen3.6-flash).
3. Updated the comment now, I hope it's more clear!

[safishamsi]

Good idea — OpenRouter slots cleanly into the existing _call_openai_compat pattern. Three things to fix:

1. Fabricated model name: qwen/qwen3.6-flash does not exist on OpenRouter. Pick a real, currently available model — qwen/qwen-2.5-coder-32b-instruct or meta-llama/llama-3.3-70b-instruct are solid cheap defaults. Check openrouter.ai/models and verify it's live before submitting.

2. No tests: Add at minimum a config-dict shape test — env key resolution (OPENROUTER_API_KEY), default model, and GRAPHIFY_OPENROUTER_MODEL override. See how test_backends.py handles the other backends.

3. Trailing whitespace: The three comment lines in the config dict end with trailing spaces — clean those up.

Fix those three and this is a clean merge.