fix(models): Correctly serialize nested schema types in LiteLlm adapter

[PaveLuchkov]

Fixes conversion of Gemini Type enums within the recursive _schema_to_dict function to ensure valid JSON Schema types ('string', 'object', etc.) are generated, resolving errors with complex tools like Google Calendar while using LiteLLM and Openrouter.

Fixes #171

Problem:
When using complex tools with nested schemas (e.g., Google Calendar insert/update) via the LiteLlm adapter, the schema generation produced invalid type representations like <Type.OBJECT: 'OBJECT'> instead of standard JSON schema types like "object". This resulted in litellm.BadRequestError: ... Provider returned error (INVALID_ARGUMENT) from the backend (e.g., Gemini via OpenRouter) due to schema validation failure.

Solution:
This PR corrects the recursive _schema_to_dict function within google.adk.models.lite_llm.py. It now uses a TYPE_LABELS dictionary lookup based on the types.Type enum value to ensure correct JSON schema type strings ('string', 'object', 'array', etc.) are generated for all nesting levels.

Testing:

1. Manual Verification: Successfully reproduced the original BadRequestError using calendar_events_update and get_weather with google/gemini-2.0-flash-001 and openai/gpt-4o-mini via LiteLLM/OpenRouter. After applying the fix, ran the same code successfully. The error is resolved.
2. Log Verification: Verified in LiteLLM debug logs that the tools payload sent to the LLM now contains the correct JSON schema types.
   * Before: ... "type": <Type.OBJECT: 'OBJECT'> ...
   * After: ... "type": "object" ...
3. Automated Tests: Ran pytest locally (using python -X utf8 -m tests/unittests/models/test_litellm.py ). Result: 36 passed, 12 skipped, 22 warnings, confirming no regressions in the tested modules.

Evidence:
The attached image shows success on the original problem from #171