ollama: add native tool calling support

[majiayu000]

PR Checklist

• Read the Contributing documentation.
• Read the Code of conduct documentation.
• Name your Pull Request title clearly, concisely, and prefixed with the name of the primarily affected package you changed according to Good commit messages (such as memory: add interfaces for X, Y or util: add whizzbang helpers).
• Check that there isn't already a PR that solves the problem the same way to avoid creating a duplicate.
• Provide a description in this PR that addresses what the PR is solving, or reference the issue that it solves (e.g. Fixes #123).
• Describes the source of new concepts.
• References existing implementations as appropriate.
• Contains test coverage for new functions.
• Passes all golangci-lint checks.

Fixes #1490

Adds native tool calling to the Ollama provider so models like Gemma 4 can use structured tool calls through the Ollama chat API.

What changed:

types.go - Added Tool, ToolFunction, ToolCall, and ToolCallFunction structs matching Ollama's API format. ToolCallFunction has custom JSON unmarshaling since Ollama returns arguments as an object but llms.FunctionCall expects a string. Added Tools field to ChatRequest and ToolCalls field to Message.

ollamallm.go - Extracted message construction into makeOllamaMessages to handle ToolCall and ToolCallResponse content parts (tool results get role=tool). Added makeOllamaTools to convert llms.Tool to ollamaclient.Tool. Response handling now converts ToolCall responses back to llms.ToolCall. Streaming is disabled when tools are present since Ollama doesn't support it.

ollama_test.go - Added TestToolCall with httprr recording covering the full flow: send message with tool defs, get tool call back, send tool result, get final text response.

Differs from the earlier #1022 attempt in a few ways: tool responses use role=tool correctly (users reported the model wasn't seeing responses in #1022), message construction is split out to keep complexity down, and tools are passed via llms.WithTools() instead of a provider-specific option.