Fix audit logging (#27): add kimi-k2.5/nemotron to tool-capable models and fix body.model resolution in invokeOllama

[GitPawanGupta]

Summary

This PR fixes two bugs related to Ollama tool calling as described in PR #27:

Bug 1: Missing tool-capable model families (src/clients/ollama-utils.js)

Added kimi-k2.5 and
emotron to the TOOL_CAPABLE_MODELS set. These newer model families support tool calling but were missing from the list, causing their tool schemas to be silently dropped.

Bug 2: invokeOllama ignores �ody.model (src/clients/databricks.js)

Fixed the model resolution logic in invokeOllama. Previously it always used config.ollama.model (global default), ignoring any model name specified in the request body. Now it correctly uses �ody.model first, falling back to config.ollama.model only when absent.

Before:
\\js
const resolvedModel = config.ollama.model;
\\

After:
\\js
const resolvedModel = (body.model && body.model.trim()) || config.ollama.model;
\\

Testing

• All 20 unit tests pass in est/ollama-tool-capable-models.test.js
• Existing model families (llama3.1, llama3.2, qwen2.5, mistral, etc.) continue to work as before
• Unrecognised models continue to be rejected