feat(models): add LiteLLM SDK model provider

[prodmanpd]

Summary

Adds LiteLLM as a model provider, so a graph can run on any provider the LiteLLM SDK supports (Anthropic, Bedrock, Vertex AI, Azure, Gemini, Groq, ...) with "model": "litellm/<provider>/<model>". LiteLLM is an optional extra (pip install scrapegraphai[litellm]) and is imported lazily, so the base install is unchanged.

Changes

• scrapegraphai/models/litellm.py: LiteLLM chat model on litellm.completion / acompletion, with drop_params=True, provider env var credentials when api_key is not set, and usage and cost in the response metadata.
• abstract_graph.py: litellm route and context window lookup.
• custom_callback.py: uses the cost a model reports when it is not in the price table.
• pyproject.toml / uv.lock: litellm>=1.94.0,<1.105.0 extra. litellm's minimums also bump openai (2.13 to 2.54), aiohttp, boto3/botocore, s3transfer, pydantic-settings and yarl in the shared lock.
• test-suite.yml: adds the new test file to the CI list.
• Example in examples/smart_scraper_graph/litellm/, README note, tests/test_litellm_model.py (23 tests).

Tests

• CI unit test list: 118 passed (23 new). The new file also passes on litellm 1.94.0 and without litellm installed.
• ruff, isort, black clean on changed files.
• Live SmartScraperGraph with a Pydantic schema on litellm/vertex_ai/gemini-2.5-flash:

llm class   : LiteLLM
model_token : 1048576 (defaulted: False)
result      : {'products': [{'name': 'Hammer', 'price': 12.5}, {'name': 'Wrench', 'price': 8.0}, {'name': 'Screwdriver set', 'price': 19.99}]}

• Live invoke and ainvoke on gpt-4.1-mini: cost split into prompt and completion, summed correctly by the callback.