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/kernels/fp8/gemm/epilogue.cuh

@@ -2,7 +2,7 @@
 // Collective epilogue: fused bias, the bf16 scatter of the fp32 accumulators
 // through the reclaimed operand shared memory, and the coalesced copy-out.
 
-#include "../common.h"
+#include "fp8/common.h"
 #include "policy.cuh"
 
 namespace astrai {

csrc/kernels/fp8/gemm/load.cuh

@@ -5,8 +5,8 @@
 // The staging invariants and the swizzle derivation live in
 // docs/developer/cuda_kernels.md.
 
-#include "../../common/cp_async.cuh"
-#include "../common.h"
+#include "common/cp_async.cuh"
+#include "fp8/common.h"
 #include "policy.cuh"
 
 namespace astrai {

csrc/kernels/fp8/gemm/mainloop.cuh

@@ -7,8 +7,8 @@
 
 #include <type_traits>
 
-#include "../../common/mma.cuh"
-#include "../common.h"
+#include "common/mma.cuh"
+#include "fp8/common.h"
 #include "load.cuh"
 #include "policy.cuh"
 

csrc/kernels/fp8/gemm/policy.cuh

@@ -5,7 +5,7 @@
 
 #include <type_traits>
 
-#include "../common.h"
+#include "fp8/common.h"
 
 namespace astrai {
 namespace fp8 {