Fix problems found when running hugging face text summariztion model on large input.

[xzdandy]

1. Pass batch_mem_size correctly to storage engines
2. Fix hugging face crash when the number of input token is large then the limit.

👋 Thanks for submitting a Pull Request to EvaDB!

🙌 We want to make contributing to EvaDB as easy and transparent as possible. Here are a few tips to get you started:

• 🔍 Search existing EvaDB PRs to see if a similar PR already exists.
• 🔗 Link this PR to a EvaDB issue to help us understand what bug fix or feature is being implemented.
• 📈 Provide before and after profiling results to help us quantify the improvement your PR provides (if applicable).

👉 Please see our ✅ Contributing Guide for more details.