feat: add OrcaRouter support for language models

[xilema2]

PR Description

# Description

This PR adds support for **OrcaRouter** (https://www.orcarouter.ai) as a new
LLM supplier in `quivr-core`. OrcaRouter is an OpenAI-compatible multi-vendor
routing gateway that exposes 150+ models from OpenAI / Anthropic / Google /
DeepSeek / xAI / MiniMax / Qwen / Kimi / etc. under a single API key, plus a
synthetic `orcarouter/auto` router that adaptively picks an upstream per
request.

Following the same pattern as the recently-merged Groq (#3633) and Gemini
(#3632) integrations, this PR:

1. Adds `ORCAROUTER = "orcarouter"` to `DefaultModelSuppliers`.
2. Registers eight flagship model entries in `LLMModelConfig._model_defaults`
   under the new supplier (auto router + one flagship per vendor). Any other
   OrcaRouter model ID is still accepted as a free-form string — the
   `LLMEndpointConfig` defaults just fall back to the generic 20k context /
   4k output budget for unrecognized models.
3. Wires up an `ORCAROUTER` branch in `LLMEndpoint.from_config` that uses
   `ChatOpenAI` against `https://api.orcarouter.ai/v1` (the SDK base URL —
   override via `LLMEndpointConfig.llm_base_url` if you point at a
   self-hosted instance). Reasoning models (`anthropic/claude-opus-4.7`,
   `openai/gpt-5*`, `deepseek-reasoner`) reject `temperature` upstream, so
   it's set to `None` for those.

### Usage

```python
from quivr_core.rag.entities.config import (
    LLMEndpointConfig,
    DefaultModelSuppliers,
)
from quivr_core.llm import LLMEndpoint

# API key from env: ORCAROUTER_API_KEY
config = LLMEndpointConfig(
    supplier=DefaultModelSuppliers.ORCAROUTER,
    model="orcarouter/auto",  # or "openai/gpt-5.5", "anthropic/claude-opus-4.7", ...
)
llm = LLMEndpoint.from_config(config)

Disclosure

I'm an engineer on the OrcaRouter team. Happy to iterate on naming, model
list, or default budgets — just let me know what fits Quivr's conventions
best.

Checklist before requesting a review

• My code follows the style guidelines of this project
• I have performed a self-review of my code
• I have commented hard-to-understand areas
• I have ideally added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged

---

Summary by cubic

Add OrcaRouter as a new LLM supplier, enabling OpenAI-compatible routing to 150+ models through a single endpoint and a synthetic orcarouter/auto router. This adds default model configs, a default base URL, and small behavior tweaks for reasoning models.

• New Features
  • Added ORCAROUTER to DefaultModelSuppliers.
  • Registered eight default model configs (including orcarouter/auto and vendor flagships); other OrcaRouter IDs still work with generic defaults.
  • Integrated via ChatOpenAI targeting https://api.orcarouter.ai/v1 by default (override with llm_base_url); sets temperature=None for reasoning models like anthropic/claude-opus-4.7, openai/gpt-5*, and deepseek-reasoner.
  • info() now reports the resolved base URL using _effective_base_url.
  • Supports ORCAROUTER_API_KEY and avoids mutating the caller’s config in from_config.

^{Written for commit dc0a6f5. Summary will update on new commits. Review in cubic}

[cubic-dev-ai]

No issues found across 3 files

_{Re-trigger cubic}

[arturict]

Could you include expected token limits or performance delta from adding Astraflow provider compared to existing providers?

[arturict]

(follow-up) For provider support changes, please include a minimal config example for one new provider to improve review clarity.