fix(harness): improve error message when model API key is missing

[anj-s]

Summary

• Implemented robust credentials validation inside the Antigravity Python harness server (harness_server.py) before entering the agent session context.
• Validates environment variables (GEMINI_API_KEY, GOOGLE_API_KEY, GOOGLE_GENAI_USE_VERTEXAI, GOOGLE_GENAI_USE_ENTERPRISE) with explicit truthiness checking (e.g. ignoring "False" strings).
• Validates programmatic credentials defined directly in configuration files, including top-level properties and nested gemini_config (supporting both shared and model-specific models.default.api_key properties).
• Cleans up standard imports (moving import os to the top of harness_server.py).
• Added pytest unit tests:
  • test_grpc_connect_missing_credentials: verifies failure response when credentials are not configured.
  • test_grpc_connect_programmatic_credentials: verifies success response when credentials are set programmatically inside the config file despite missing environment variables.
• Updated mock_config fixture to inject mock env keys by default, ensuring existing success tests pass cleanly in clean CI environments.

Type of Change

• Bug fix
• Refactor

Related Issues

Closes #63

Test Plan

• Python tests pass: pytest python/antigravity/harness_server_test.py
• Go integration tests pass: go test ./...