[XPU] fix xpu_wait for bkcl_send

[lj970926]

PR Category

Custom Device

PR Types

Bug fixes

Description

Because bkcl_send and bkcl_recv use a seperate stream for comm and would not wait original stream, we add a forced xpu_wait before send/recv

[paddle-bot[bot]]

你的PR提交成功，感谢你对开源项目的贡献! 请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。 Your PR has been submitted. Thanks for your contribution! Please wait for the result of CI firstly. See Paddle CI Manual for details.