fix: keep embedding string batches separate

[he-yufeng]

Summary

• keep pure list[str] inputs batched for Gemini embedding 2 on the Developer API path
• preserve the existing multimodal behavior where mixed text/image parts are combined into one content item
• cover both sync and async embed_content() argument handling

Fixes #2523.

Verification

$env:PYTHONPATH='.'; $env:GOOGLE_GENAI_REPLAYS_DIRECTORY="$env:TEMP\google-genai-replays"; python -m pytest google\genai\tests\models\test_embed_content.py -k "text_list_stays_batched or mixed_content_still_combines_parts" -q
python -m py_compile google\genai\models.py google\genai\tests\models\test_embed_content.py
python -m ruff check --select E9,F63,F7,F82 google\genai\models.py google\genai\tests\models\test_embed_content.py
git diff --check

I also checked the converted Developer API request shape directly: contents=["s1", "s2"] now produces two batchEmbedContents requests instead of one multi-part request.