feat: Update tests & examples to use granite4

[planetf1]

Migration to Granite 4.0 Models 🚀

Why this matters:
The granite-3.x model family is approaching its End of Life (EOL). Default configurations, examples, and tests must be migrated to the granite-4.0 family.

What needs to be done:
Identify all usages of Granite v3 (and earlier) and update them to their v4 equivalents.

Observations:
There maybe some confusion in mellea/backends/model_ids.py. The constant IBM_GRANITE_4_MICRO_3B is mapped to ibm/granite-4-h-small. However, ibm/granite-4-h-small is a "Hybrid Small" model (larger and more capable than "Micro"). This naming mismatch should be corrected during the migration.
Additionally, explicit Granite 4 usages were found in:

• test/stdlib/components/intrinsic/test_rag.py: Uses ibm-granite/granite-4.0-micro directly.
• test/conftest.py: References granite4:micro in docstrings for Ollama checks.
  These should be harmonized with the new default constants.

Action Items:

• Model IDs: Address the observation above. Create a new constant (e.g., IBM_GRANITE_4_HYBRID_SMALL) for ibm/granite-4-h-small and ensure IBM_GRANITE_4_MICRO_3B points to a true micro model (if available) or is deprecated/renamed.
• Audit Usages: Check all usages of IBM_GRANITE_4_MICRO_3B to determine if they intended to use the Micro model or the Hybrid Small model (since the constant currently points to Hybrid Small on Watsonx).
• Backend Defaults: Update WatsonxAIBackend default model to ibm/granite-4-h-small (using the new correct constant).
  • Note: The WatsonxAIBackend itself is deprecated in favor of LiteLLM (as noted in the codebase). However, the default should still be fixed while it remains in use.
• Documentation: Update docs/tutorial.md and other guides to reference Granite 4.
• Examples: Ensure all examples in docs/examples/ run out-of-the-box with Granite 4.
• Tests: Verify uv run pytest passes with the new defaults.

Note that this is also the cause of the deprecation warning seen during testing LifecycleWarning: Model 'ibm/granite-3-3-8b-instruct' is in deprecated state... alternative models: ibm/granite-4-h-small.

Affected Files (Preliminary Scan):
References to "granite" were found in the following files.:

Configuration & Core

• pyproject.toml
• mellea/backends/model_ids.py
• mellea/backends/watsonx.py
• mellea/backends/ollama.py
• mellea/backends/huggingface.py
• mellea/backends/litellm.py
• mellea/backends/openai.py

Documentation & Examples

• docs/tutorial.md
• docs/AGENTS_TEMPLATE.md
• docs/alora.md
• docs/dev/mellea_library.md
• docs/examples/aLora/101_example.py
• docs/examples/notebooks/model_options_example.ipynb
• docs/examples/notebooks/simple_email.ipynb
• docs/examples/instruct_validate_repair/101_email_with_validate.py (and others in docs/examples/)
• docs/kv_smash/hf_example.py

Tests

• test/backends/test_watsonx.py
• test/backends/test_ollama.py
• test/backends/test_huggingface.py
• test/conftest.py

I can start working through this - or ok if we want to split up. Does the scope make sense?
Do we want to reain any v3 tests? I presume not

[planetf1]

Starting to work on this. I may not be able to test everything locally, so will propose an update someone with a larger machine can test. There's a few parts to this so will continue into next week.