fix(gemini): group parallel tool responses

[Zbhbb]

Description

Gemini expects responses to parallel function calls in one user Content. Previously, the formatter emitted one Content per ToolResultBlock, which can split a valid parallel tool-response batch.

This change buffers tool-result parts while converting one Msg, then emits one user Content containing the ordered function responses.

Changes

• Group tool-result parts from a single message into one user Content.
• Preserve the original tool-result order.
• Add a regression test for two tool results in one message.

Testing

• mvn -pl agentscope-extensions/agentscope-extensions-model/agentscope-extensions-model-gemini -am -q -Dtest=GeminiMessageConverterTest -Dsurefire.failIfNoSpecifiedTests=false test
• mvn -pl agentscope-extensions/agentscope-extensions-model/agentscope-extensions-model-gemini -am -q test
• mvn -pl agentscope-extensions/agentscope-extensions-model/agentscope-extensions-model-gemini -q spotless:check
• git diff --check

Closes #2351

Checklist

• Code has been formatted with Spotless.
• Relevant module tests are passing.
• Code is ready for review.

[CLAassistant]

All committers have signed the CLA.

[Copilot]

Pull request overview

This PR fixes Gemini function-calling formatting by ensuring multiple tool results produced in a single tool message are emitted as a single Content(role="user"), matching Gemini’s contract for parallel function responses.

Changes:

• Buffer ToolResultBlock conversions per Msg and emit one grouped user Content containing all FunctionResponse parts in original order.
• Preserve existing conversion behavior for non-tool blocks while keeping tool responses separated into their own Content.
• Add a regression test covering two tool results in one message.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
.../GeminiMessageConverter.java	Groups tool-result parts from a single Msg into one user Content instead of emitting one per ToolResultBlock.
.../GeminiMessageConverterTest.java	Adds a test asserting two tool results become one user Content with two FunctionResponse parts in order.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[oss-maintainer]

LGTM. Looks good.

---

Automated review by "github-manager-bot"

[AgentScopeJavaBot]

🤖 AI Review

This PR fixes a Gemini API compatibility issue: when the model issues parallel function calls, the framework previously emitted each ToolResultBlock as a separate user-role Content object. The Gemini API requires all function_response parts for parallel calls from a single model turn to be sent in one user turn. The fix collects all function_response parts from a single Msg into one user-role Content.

(inline comments could not be attached — line numbers fell outside PR hunks. See archived report.)

[AgentScopeJavaBot]

🤖 AI Review

This PR fixes a Gemini API compatibility issue: when the model issues parallel function calls, the framework previously emitted each ToolResultBlock as a separate user-role Content object. The Gemini API requires all function_response parts for parallel calls from a single model turn to be sent in one user turn. The fix collects all function_response parts from a single Msg into one user-role Content.

(inline comments could not be attached — line numbers fell outside PR hunks. See archived report.)

[AgentScopeJavaBot]

🤖 AI Review

This PR fixes a Gemini API compatibility issue: when the model issues parallel function calls, the framework previously emitted each ToolResultBlock as a separate user-role Content object. The Gemini API requires all function_response parts for parallel calls from a single model turn to be sent in one user turn. The fix collects all function_response parts from a single Msg into one user-role Content.

(inline comments could not be attached — line numbers fell outside PR hunks. See archived report.)