Feature/bugfix

[jxt1234]

1. 修正常量内存优化导致的部分模型使用GPU运行出错的问题
2. 修正 scatterelement 处理 shape = 0 的情况下的问题
3. 修正 Vulkan Softmax 算子对于输入为负的大数时，计算出现 nan 的问题
4. 修正 ROIAlign 算子在 fp16 / bf16 模式下可能内存越界的问题

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

xiaying seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}