refactor(pr66): trim commit-narration comments

/simplify found 4 comment blocks that narrate the rename rationale
rather than encode load-bearing contracts:

- `kernel_custom.h` forward-decl — compress build-system detail
  (`no_workspace_kernel`, `ascendc_library()`) to one line, keep only
  the ABI contract (`aclrtlaunch_<Entry>` is generated by AscendC from
  `op_kernel/`).
- `op_host/<op>.cpp` `EXEC_KERNEL_CMD` — drop "Parameter order follows
  the base class: inputs, attributes, outputs."; the signature itself
  is self-evident.
- `op_kernel/<op>.cpp` kernel entry — drop "Parameters follow the C2
  convention ..." and "`aclrtlaunch_AddRmsNorm` matches the base
  `AddRmsNorm` class name"; these are commit-message material, not
  comments.

Author: zhangyue

src/ascend/add_rms_norm/kernel_custom.h

@@ -14,11 +14,8 @@
 #include "base/add_rms_norm.h"
 #include "operator.h"
 
-// Forward-declare the generated AscendC kernel launch function.  This
-// symbol is provided by the `no_workspace_kernel` static library built
-// from `ascend/custom/add_rms_norm/op_kernel/add_rms_norm.cpp` via
-// `ascendc_library()`; the `aclrtlaunch_` prefix is prepended by the
-// AscendC toolchain to the kernel entry's `extern "C"` name.
+// Forward-declare the `aclrtlaunch_AddRmsNorm` launch symbol defined
+// by the AscendC toolchain from `custom/add_rms_norm/op_kernel/`.
 extern "C" uint32_t aclrtlaunch_AddRmsNorm(
     uint32_t block_dim, void* stream, void* input, void* residual, void* weight,
     int64_t total_rows, int64_t dim_length, int64_t dim_length_align,

src/ascend/custom/add_rms_norm/op_host/add_rms_norm.cpp

@@ -108,7 +108,6 @@ std::vector<at::Tensor> AddRmsNorm(const at::Tensor& x1, const at::Tensor& x2,
   // The first arg `AddRmsNorm` is the AscendC kernel entry-point name — it
   // must match the `__global__ __aicore__ void AddRmsNorm(...)` definition
   // in `op_kernel/` and the generated `aclrtlaunch_AddRmsNorm.h` header.
-  // Parameter order follows the base class: inputs, attributes, outputs.
   EXEC_KERNEL_CMD(AddRmsNorm, block_dim, kernel_input1, kernel_input2,
                   weight_float, total_rows, dim_length, dim_length_align,
                   former_num, former_length, tail_length, eps_float,

src/ascend/custom/add_rms_norm/op_kernel/add_rms_norm.cpp

@@ -247,11 +247,6 @@ class KernelAddRmsNorm {
 // by the host launcher.  fp16 and bf16 both have `sizeof == 2` but need
 // distinct numeric paths, so dispatch is on the `DataType` tag rather
 // than the byte size.
-//
-// Parameters follow the C2 convention: inputs first, attributes between,
-// outputs last.  The kernel symbol is prefixed with `aclrtlaunch_` by the
-// `AscendC` toolchain, yielding `aclrtlaunch_AddRmsNorm` which matches the
-// base `AddRmsNorm` class name.
 extern "C" __global__ __aicore__ void AddRmsNorm(
     GM_ADDR input, GM_ADDR residual, GM_ADDR weight, int64_t total_rows,
     int64_t dim_length, int64_t dim_length_align, int64_t former_num,

src/ascend/custom/rms_norm/op_host/rms_norm.cpp

@@ -97,7 +97,6 @@ at::Tensor RmsNorm(const at::Tensor& input, const at::Tensor& weight,
   // The first arg `RmsNorm` is the AscendC kernel entry-point name — it
   // must match the `__global__ __aicore__ void RmsNorm(...)` definition in
   // `op_kernel/` and the generated `aclrtlaunch_RmsNorm.h` header.
-  // Parameter order follows the base class: inputs, attributes, outputs.
   EXEC_KERNEL_CMD(RmsNorm, block_dim, kernel_input, weight_float, total_rows,
                   dim_length, dim_length_align, former_num, former_length,
                   tail_length, eps_float, dtype_size_val, kernel_output);

src/ascend/custom/rms_norm/op_kernel/rms_norm.cpp

@@ -205,11 +205,6 @@ class KernelRmsNorm {
 // by the host launcher.  fp16 and bf16 both have `sizeof == 2` but need
 // distinct numeric paths, so dispatch is on the `DataType` tag rather
 // than the byte size.
-//
-// Parameters follow the C2 convention: inputs first, attributes between,
-// outputs last.  The kernel symbol is prefixed with `aclrtlaunch_` by the
-// `AscendC` toolchain, yielding `aclrtlaunch_RmsNorm` which matches the
-// base `RmsNorm` class name.
 extern "C" __global__ __aicore__ void RmsNorm(
     GM_ADDR input, GM_ADDR weight, int64_t total_rows, int64_t dim_length,
     int64_t dim_length_align, int64_t former_num, int64_t former_length,

src/ascend/rms_norm/kernel_custom.h

@@ -15,11 +15,8 @@
 #include "base/rms_norm.h"
 #include "operator.h"
 
-// Forward-declare the generated AscendC kernel launch function.  This
-// symbol is provided by the `no_workspace_kernel` static library built
-// from `ascend/custom/rms_norm/op_kernel/rms_norm.cpp` via
-// `ascendc_library()`; the `aclrtlaunch_` prefix is prepended by the
-// AscendC toolchain to the kernel entry's `extern "C"` name.
+// Forward-declare the `aclrtlaunch_RmsNorm` launch symbol defined by
+// the AscendC toolchain from `custom/rms_norm/op_kernel/`.
 extern "C" uint32_t aclrtlaunch_RmsNorm(
     uint32_t block_dim, void* stream, void* input, void* weight,
     int64_t total_rows, int64_t dim_length, int64_t dim_length_align,