Clean up dead code from legacy MMQ implementation

[Isotr0py]

• Clean up dead code remained from [1/N] Update MMQ kernel catch up upstream llama.cpp implementation #5

[Copilot]

Pull Request Overview

This PR cleans up dead code from a legacy MMQ (mul_mat_q) implementation. The changes involve removing unused function definitions, macro definitions, and associated tile allocation/loading functions that are no longer needed after the implementation was refactored in PR #5.

Key changes:

• Removed numerous MMQ-related macro definitions (VDR_*_MMQ)
• Moved core MMQ implementation functions to a dedicated header file
• Eliminated dead template functions for memory allocation and tile loading

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
hf-kernels/ggml-kernels/ggml/vecdotq.cuh	Removed legacy MMQ macro definitions and dead implementation functions while preserving MMVQ functionality
hf-kernels/ggml-kernels/ggml/mmq.cuh	Added consolidated MMQ implementation functions and macro definitions to new dedicated header

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}