Place ckProfiler headers into include/ck/profiler

[poyenc]

Currently we put headers into include/ck/xxxxx sub-directories except ckProfiler

$ tree library/include/ -L 3
library/include/
└── ck
    └── library
        ├── reference_tensor_operation
        ├── tensor_operation_instance
        └── utility

$ tree profiler/include/ -L 2
profiler/include/
├── data_type_enum_helper.hpp
├── data_type_enum.hpp
├── profile_batched_gemm_add_relu_gemm_add_impl.hpp
├── profile_batched_gemm_gemm_impl.hpp
├── profile_batched_gemm_impl.hpp
├── profile_batched_gemm_masking_scale_softmax_gemm_permute_impl.hpp
├── profile_batched_gemm_reduce_impl.hpp
├── profile_batched_gemm_softmax_gemm_impl.hpp
├── profile_conv_bwd_data_impl.hpp
├── profile_conv_bwd_weight_impl.hpp
├── profile_conv_fwd_bias_relu_add_impl.hpp
├── profile_conv_fwd_bias_relu_impl.hpp
├── profile_conv_fwd_impl.hpp
├── profile_convnd_bwd_data_impl.hpp
├── profile_convnd_bwd_weight_impl.hpp
├── profile_gemm_add_add_fastgelu_impl.hpp
├── profile_gemm_bias_add_reduce_impl.hpp
├── profile_gemm_bilinear_impl.hpp
├── profile_gemm_impl.hpp
├── profile_gemm_reduce_impl.hpp
├── profile_gemm_splitk_impl.hpp
├── profile_grouped_conv_fwd_impl.hpp
├── profile_grouped_gemm_impl.hpp
├── profile_groupnorm_impl.hpp
├── profile_layernorm_impl.hpp
├── profile_normalization_impl.hpp
└── profile_reduce_impl.hpp

This leads to inconsistent include_directories() usage and special include directives

// Instead of doing this...
#include "profiler/include/profile_gemm_impl.hpp"

// Use include paths like below
#include "ck/profiler/profile_gemm_impl.hpp"

I suggest that moving ckProfiler headers into profiler/include/ck/profiler sub-folder.