Expand tool calling test coverage across providers and error paths

[Kamilbenkirane]

Priority: High
Related: #199

Current test coverage has significant gaps:

Provider coverage: Only 4/13 providers tested (Anthropic, OpenAI, Google, xAI). Groq, Mistral, Moonshot, DeepSeek, HuggingFace, Ollama have zero tool tests.

Missing test categories:

• Unit tests for tool parameter mappers (schema transformation correctness)
• Error paths: unsupported tool on provider, invalid tool schema, tool on non-supporting model
• Raw passthrough tools
• Streaming tool call aggregation (mock SSE events)
• Multiple concurrent tool calls in one response
• tools=[] and tools=None behavior
• WebSearch configuration options (allowed_domains, blocked_domains, max_uses)
• Multi-turn with multiple tool calls

Test quality: Integration tests are mostly "does it not crash" — no semantic assertions on tool call arguments or result content.

Files:

• tests/integration_tests/text/test_tools.py
• tests/unit_tests/test_client.py
• tests/unit_tests/test_constraints.py

[Kamilbenkirane]

Test coverage expansion is ongoing work, not a blocker for the tools PR.