fix(ai): attach user media to Gemini requests

[ling-senpeng13]

Problem

Images passed as input via the media path never reached Gemini models — the model received a text-only message. Same bug as #1238 (Anthropic), in the Gemini converter:

case USER -> {
    UserMessage userMsg = (UserMessage) msg;
    contents.add(new GeminiApi.Content(
        "user", List.of(GeminiApi.Part.text(userMsg.getText()))));   // getMedia() dropped
}

LLMHelper.getMessage() builds a Spring AI UserMessage with media (bytes), but GeminiChatModel.convertMessage() only forwarded getText().

Fix

• GeminiChatModel.convertMessage(): when a UserMessage carries media, build text + one inline-data Part per media item (base64), mirroring the OpenAI path and the fix(ai): attach user media to Anthropic requests #1238 Anthropic fix. Text-only messages are unchanged.
• GeminiApi: add a Part.inlineData(mimeType, base64Data) factory (the InlineData record already existed; it just had no request-side factory).

Test

Adds GeminiChatModelMediaTest, which mocks the API, captures the contents passed to generateContent, and asserts the user content carries an inline image part (inlineData) with mimeType=image/png and the verbatim base64 payload — plus the accompanying text part.

Validated both directions:

• Before the fix: FAILS — "inline image part missing — media was dropped".
• After the fix: PASSES.

./gradlew :conductor-ai:test --tests "org.conductoross.conductor.ai.providers.gemini.GeminiChatModel*"
BUILD SUCCESSFUL

Scope

Gemini provider only; other providers untouched. convertMessage is the single message-building path for Gemini, so all Gemini requests are covered. Companion to #1238 (Anthropic) — the two custom converters that dropped media.

🤖 Generated with Claude Code