[onert] Support Multiple Model Files for Bulk Op

[batcheu]

What

Currently, the Bulk Op supports processing only a single model file. This task aims to modify the implementation to enable reading and processing multiple model files simultaneously.

Why

Processing multiple files together can reduce overhead and improve overall performance by minimizing repeated operations.
It is also expected that the introduction of a new API for npu-engine will improve memory usage by reusing model buffers.

[glistening]

Do you mean to run several tvn simultaneously in one Bulk?
How the relationship between multiple tvns in Bulk are described?

(ADD) I've found #16281.

@batcheu seems to assume the models are the sequence of tvns like decoder layers in Transformers.