Split prefill/decode within single mxr

[turneram]

Motivation

Add split prefill/decode support for dynamic-sequence ONNX models. This enables efficient static specialization of decode (sequence_length=1) and prefill (sequence_length=max) workloads while preserving a single runtime-facing program.

Technical Details

• Added ONNX parser options for split prefill/decode and symbolic shapes.
• Generates static decode and prefill submodules behind a runtime select_module dispatcher.
• Exposed the options through the C++, Python file/buffer, and driver interfaces.
• Preserved symbolic dimensions through optimization and serialization.
• Improved evaluation and output allocation for fused multi-output tuple parameters, including loop and scan outputs.
• Added reference, GPU, serialization, GQA, multi-I/O, driver, Python, and unsupported-length validation.
• Added developer documentation covering architecture, usage, and validation.

Changelog Category

Add a CHANGELOG.md entry for any option other than Not Applicable

• Added: New functionality.
• Changed: Changes to existing functionality.
• Removed: Functionality or support that has been removed. (Compared to a previous release)
• Optimized: Component performance that has been optimized or improved.
• Resolved Issues: Known issues from a previous version that have been resolved.
• Not Applicable: This PR is not to be included in the changelog.

Follow the LLVM AI Tool Use Policy for contributions using AI.