Add support for cuBLASLt functions used by llm.c

[zluda-violet]

Depends on #510

[Copilot]

Pull Request Overview

This PR adds support for cuBLASLt functions used by llm.c by implementing comprehensive cuBLAS Lite functionality. It expands the ZLUDA cuBLAS Lite implementation from basic handle management to full matrix multiplication operations including descriptors, preferences, layouts, and algorithm heuristics.

• Implements core cuBLASLt functions for matrix operations (matmul, descriptors, preferences, layouts)
• Adds comprehensive type conversions between CUDA and HIP data types and enums
• Integrates hipblaslt-sys dependency for HIP BLAS Lite backend functionality

Reviewed Changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
zluda_common/src/lib.rs	Adds type conversions for cuBLASLt data types and enums to HIP equivalents
zluda_common/Cargo.toml	Adds hipblaslt-sys dependency
zluda_blaslt/src/lib.rs	Expands implemented function list to include matmul operations
zluda_blaslt/src/impl.rs	Implements cuBLASLt function handlers with HIP backend calls
zluda_blaslt/Cargo.toml	Adds hip_runtime-sys and hipblaslt-sys dependencies

---

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

[vosen]

This whole wrapping of hipblasLtMatmulDesc_t, hipblasLtMatmulPreference_t and hipblasLtMatrixLayout_t seems unnecessary to me, if we wrapped those types with some extra data, but as it is now it's better to map them 1:1

[zluda-violet]

Makes sense, done.