fix: reject multiple response candidates

[svfat]

Summary

Closes #6518.

• reject candidate_count > 1 before sending model request
• apply same single-candidate contract to Gemini, LiteLLM and Apigee paths
• add regression tests for request validation and existing first-candidate response conversion

Why

ADK exposes one model result through LlmResponse, but behavior for
candidate_count > 1 was inconsistent:

• LiteLLM ignored the option and requested one choice
• native Gemini requested multiple candidates and ADK kept only first one
• Apigee requested multiple choices and kept only first one

This can give false expectation that multiple results are available. It can
also waste billed candidate tokens when extra outputs are generated and then
discarded.

Complete multi-candidate support would require change to LlmResponse and
downstream event/streaming APIs. This patch keeps current single-candidate
contract and returns clear error instead of silently ignoring or discarding
results.

Testing

• pytest -q tests/unittests/models/test_llm_response.py tests/unittests/models/test_google_llm.py tests/unittests/models/test_litellm.py tests/unittests/models/test_completions_http_client.py
  • 471 passed
• focused tests on Python 3.10, 3.11, 3.12, 3.13 and 3.14
• pre-commit hooks
• Talisman