fix: merge streamed model history turns

[he-yufeng]

Fixes #2508.

Summary

When a streaming response yields multiple valid model Content chunks, the chat currently keeps those chunks as consecutive model turns in curated history. That history is reused for later requests, so the next request can include back-to-back model turns instead of a single model turn.

This change keeps comprehensive history unchanged, but normalizes curated history by merging adjacent model chunks into one model turn. That preserves the original parts while keeping request history in the expected user/model turn shape.

Tests

python -m pytest google\genai\tests\chats\test_get_history.py -q
python -m pytest google\genai\tests\chats\test_validate_response.py -q
python -m mypy google\genai\chats.py
python -m py_compile google\genai\chats.py google\genai\tests\chats\test_get_history.py
git diff --check

[he-yufeng]

Rebased this branch onto current main and pushed the updated head 77160fd.

Validation:

• python -m pytest google\genai\tests\chats\test_get_history.py -q -k "stream_merges_model_chunks_for_curated_history" -> passed, 1 passed
• python -m py_compile google\genai\chats.py google\genai\tests\chats\test_get_history.py -> passed
• git diff --check upstream/main..HEAD -> passed

[he-yufeng]

Rebased onto current main. Upstream's fe6118e reworked AFC in chat streams and deleted the two AFC history fixtures along with their tests, so this branch drops them too and keeps only the new split-output fixture and the curated-history regression test. The merge behavior is unchanged; the rebased suite passes locally (18 passed, including the new test).