Streamed Multi-Head Attention for Altera Backend

[bugracyln]

Description

📝 Please include a summary of the change.

• Implementation of MHA for io_stream version.
• WARNING: This change relies on pr Rename oneAPI backend to Altera, remove oneAPI compatibility #1536, so it should be merged after it.

Type of change

For a new feature or function, please create an issue first to discuss it
with us before submitting a pull request.

Note: Please delete options that are not relevant.

• Bug fix (non-breaking change that fixes an issue)
• Documentation update
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• A new research paper code implementation
• Other (Specify)

Tests

📝 Please describe the tests that you ran to verify your changes.

• The module has been tested extensively with various test scripts, both for accuracy testing by using the bridge to use .predict() feature to compare with the actual front-end-produced multi-head attention layer. This module has also been tested on an actual FPGA inside a small LLM architecture and was confirmed to provide correct results.

• A simple demo is provided for experimentation. (see streamed_MHA_test_script.py)

Checklist

• I have read the guidelines for contributing.
• I have commented my code, particularly in hard-to-understand areas.
• I have made corresponding changes to the documentation.
• My changes generate no new warnings.
• I have installed and run pre-commit on the files I edited or added.
• I have added tests that prove my fix is effective or that my feature works.