[Improvement] Management of Small Files and Data Acceleration in AI Scenarios

### What would you like to be improved?

In the AI scenario, managing related files with FileSet may have a scenario with too many small files. How does Gravitino consider the need to merge small files? In addition, how to accelerate data? And is there a consideration for implementing cold storage for related data?

### How should we improve?

_No response_

Start by reviewing how FileSet manages related files and how the project handles data acceleration or cold storage; the issue names no files, tests, or entry points. Before implementation, clarify the desired small-file merge, acceleration, and cold-storage behavior, then define completion criteria for those requirements.

Written by the indexing model from the issue text.