Bug: Wrong number of tool calls using evals-cli

[beaufortfrancois]

Try I'd like a small Pesto pizza for https://googlechromelabs.github.io/webmcp-tools/demos/pizza-maker/ using the evals-cli tool.

Vercel backend gives me the following tool calls:

1. call: {"type":"tool-call","toolCallId":"Cdd9DKnzVP2PiLC3","toolName":"set_pizza_style","input":{"style":"Pesto"},"providerMetadata":{"google":{"thoughtSignature":"..."}}}
2. call: {"type":"tool-call","toolCallId":"oJlgIdNgaY3BteLG","toolName":"set_pizza_style","input":{"style":"Pesto"},"providerMetadata":{"google":{"thoughtSignature":"..."}}}
3. call: {"type":"tool-call","toolCallId":"EdJOFFtzL97582Hv","toolName":"set_pizza_size","input":{"size":"Small"}}

Gemini backend only returns the first one:

'{"parts":[{"functionCall":{"name":"set_pizza_style","args":{"style":"Pesto"}},"thoughtSignature":"..."}],"role":"model"}'

Using the Model Context Inspector extension, I get the 2 appropriate tool calls:

User prompt: "I'd like a small Pesto pizza"
AI calling tool "set_pizza_size" with {"size":"Small"}
Tool "set_pizza_size" result: Set pizza size to Small.
AI calling tool "set_pizza_style" with {"style":"Pesto"}
Tool "set_pizza_style" result: Changed pizza style to Pesto
AI result: I have a small pesto pizza for you!

[Kulikowski]

Support added in #47