test: add circular access in TensorBase::assign (AIV-802)

[Vope17]

Implement circular access with start_position in
TensorBase::assign() for the INT8 data type.

Description

Optimize handling of circular data to avoid extra copy costs. For time-sequence data like
[1, 2, 3, 4, 5],
when new data (6) arrives, instead of shifting all elements to
[2, 3, 4, 5, 6],
we update start_position to 1 (value 2) represent the start index:
[6, 2, 3, 4, 5].
The storage method for new data uses circular indexing as shown below:
current_idx = (current_idx + 1) % WINDOW_SIZE;

Related

Testing

Tested on esp-idf-v5.3.1 with arch linux x86_64 on esp32s3

---

Checklist

Before submitting a Pull Request, please ensure the following:

• 🚨 This PR does not introduce breaking changes.
• All CI checks (GH Actions) pass.
• Documentation is updated as needed.
• Tests are updated or added as necessary.
• Code is well-commented, especially in complex areas.
• Git history is clean — commits are squashed to the minimum necessary.