[codex] Expand streaming aggregate support

[ashione]

What changed

• expand streaming aggregates from single-output support to shared multi-aggregate specs across the DataFrame API and stream SQL
• add MIN, MAX, AVG, multi-aggregate stream SQL output, and HAVING alias support
• extend actor hot-path eligibility to COUNT(*) while keeping non-hot-path and multi-aggregate shapes on local execution with explicit fallback reasons
• align planner explain eligibility with runtime constraints and add Python SQL regression coverage for the new behavior
• update English and Chinese docs plus streaming runtime design notes to reflect the supported boundary

Why

The previous implementation only handled a narrow single-aggregate shape and could over-report actor eligibility in explain output. This change widens the supported streaming aggregate surface while keeping planner/runtime behavior aligned and covered by regression tests.

Impact

Users can now run richer streaming aggregate queries through both C++ and Python entrypoints, including combined COUNT/SUM/MIN/MAX/AVG projections and HAVING on aggregate aliases. Actor acceleration remains intentionally limited to the supported window_start,key + SUM(value) and window_start,key + COUNT(*) shapes.

Validation

• bazel test //:stream_runtime_test
• bazel test //:stream_strategy_explain_test
• bazel test //:core_regression //:experimental_regression
• VELARIA_PYTHON_BIN=/Users/bytedance/.local/share/uv/python/cpython-3.13.5-macos-aarch64-none/bin/python3.13 bazel test //python_api:streaming_v05_test