fix: handle multiple tool calls in tools examples

[Krishnachaitanyakc]

Summary

• Fixes the tools.py and async-tools.py examples so that all tool call results are sent back to the model, not just the last one
• Previously, the examples iterated over tool calls but only appended a single tool message after the loop using the last value of output, which meant earlier tool call results were silently discarded
• Each tool result is now appended as a separate message inside the loop, consistent with the pattern already used in multi-tool.py and gpt-oss-tools.py

Fixes #476

Changes

• examples/tools.py: Move messages.append(response.message) before the loop; append each tool result inside the loop
• examples/async-tools.py: Same fix applied to the async variant

Test plan

• ruff check passes on both changed files
• Python syntax validation passes
• Manual verification: run tools.py with a prompt that triggers multiple tool calls (e.g., "What is three plus one, then subtract two?") and confirm all results are sent to the model

[Pawansingh3889]

Good catch — I ran into this exact issue when building a multi-tool agent with Ollama. The original code only appended the last tool result because the loop variable got overwritten on each iteration.

The fix correctly:

1. Appends the assistant message (with tool_calls) before iterating
2. Appends each tool result inside the loop as a separate message
3. Adds an "Function not found" fallback for missing functions

One minor thought: the fallback string "Function not found" could be more descriptive — something like f"Error: function {tool.function.name} not available" would help the model self-correct in the next turn.

Tested the same pattern in my own project and it works correctly. LGTM.