Replace YAML operations with C libyaml backend

Objectives

Reduce the runtime of Tensile applications and any downstream build chains by replacing the standard PyYAML Loader and Dumper with the C libyaml backend.

Testing:

TensileCreateLibrary

When timing the execution of TensileCreateLibrary on gfx90a architectures, the following differences were noted:

Before (with SafeLoader): real 6m26s
After (with CSafeLoader): real 5m13s
Result: 19.0% runtime reduction

Command issued:

$ time Tensile/bin/TensileCreateLibrary ${rocblas_root}/library/src/blas3/Tensile/Logic/asm_full/aldebaran build_profiler HIP --merge-files --separate-architecture --lazy-library-loading --no-short-file-names --no-library-print-debug --code-object-version=default --cxx-compiler=hipcc --jobs=64 --library-format=msgpack --architecture=gfx90a

rocBLAS

When building rocBLAS, the following differences were noted:

Before (with SafeLoader): real 53m46s
After (with CSafeLoader): real 51m57s
Result: 3.4% runtime reduction

Command issued:

$ ./install.sh

ROCm / Tensile

Replace YAML operations with C libyaml backend #1937