Fix inference bugs in Qwen2.5 Omni (#37701)
* Init `SinusoidsPositionEmbedding` with float to avoid precision problem
* fix hidden_state for talker
* Update modular_qwen2_5_omni.py
* Move hidden processing out from thinker
* fixup
---------
Co-authored-by: lvyuanjun.lyj <lvyuanjun.lyj@alibaba-inc.com>
