refactor: root kernel includes at csrc/kernels

- add the kernels directory to CMake target include paths and drop all ../-relative includes in kernel sources
- reference shared primitives as common/*.cuh and the fp8 type header as fp8/common.h
- update standalone test nvcc commands in file headers and cuda_kernels.md to -I csrc/kernels

csrc/tests/attn_paged_test.cu

@@ -1,5 +1,5 @@
 // Compile:
-//   nvcc -I csrc -arch=sm_89 -O3 --use_fast_math --ptxas-options=-O3 \
+//   nvcc -I csrc/kernels -arch=sm_89 -O3 --use_fast_math --ptxas-options=-O3 \
 //        --extra-device-vectorization -Xcompiler -fopenmp \
 //        csrc/tests/attn_paged_test.cu \
 //        -o /tmp/test_paged && /tmp/test_paged
@@ -7,7 +7,7 @@
 #include <cstring>
 #include <vector>
 #include "test_utils.cuh"
-#include "../kernels/attention/dispatchers.cuh"
+#include "attention/dispatchers.cuh"
 
 using namespace astrai::attention;
 

csrc/tests/attn_test.cu

@@ -1,13 +1,13 @@
 /*
 Pure-C test — uses shared dispatcher.  Combines the decode (split-KV) and
 prefill (split-Q) correctness checks + benchmarks into one binary.
-nvcc -I csrc -arch=sm_89 -O3 \
+nvcc -I csrc/kernels -arch=sm_89 -O3 \
     --use_fast_math --ptxas-options=-O3 --extra-device-vectorization \
     -Xcompiler -fopenmp csrc/tests/attn_test.cu -o test && ./test
 */
 
 #include "test_utils.cuh"
-#include "../kernels/attention/dispatchers.cuh"
+#include "attention/dispatchers.cuh"
 
 using namespace astrai::attention;
 

csrc/tests/fp8_test.cu

@@ -6,7 +6,7 @@ Part 1 exercises one bf16 -> fp8 -> mma.sync m16n8k32 instruction pair
 Part 2 checks launch_fp8_gemm across all four operand layouts, both K
 tiles, and ragged shapes against an fp32 CPU reference.
 
-nvcc -I csrc -arch=sm_89 -std=c++17 -O3 csrc/tests/fp8_test.cu -o /tmp/fp8_test \
+nvcc -I csrc/kernels -arch=sm_89 -std=c++17 -O3 csrc/tests/fp8_test.cu -o /tmp/fp8_test \
     && /tmp/fp8_test
 */
 
@@ -21,8 +21,8 @@ nvcc -I csrc -arch=sm_89 -std=c++17 -O3 csrc/tests/fp8_test.cu -o /tmp/fp8_test
 #include <type_traits>
 #include <vector>
 
-#include "../kernels/common/mma.cuh"
-#include "../kernels/fp8/gemm.cuh"
+#include "common/mma.cuh"
+#include "fp8/gemm.cuh"
 
 using namespace astrai::fp8;