Rename candle's moe_gemm_wmma -> candle_moe_gemm_wmma to avoid symbol collision with mistralrs-core

Both candle-kernels and mistralrs-core define an extern "C" void moe_gemm_wmma
kernel with the same signature but incompatible implementations. When both
crates are linked into the same binary (e.g. spiced with cuda + models
features), ld.lld rejects the duplicate symbol.

Prefix the candle side with 'candle_' so both can coexist; the mistralrs-core
copy keeps the original name and call sites.

Author: lukekim

candle-kernels/src/ffi.rs

@@ -2,7 +2,7 @@ use core::ffi::c_void;
 #[allow(dead_code)]
 extern "C" {
     // for unquntized models
-    pub fn moe_gemm_wmma(
+    pub fn candle_moe_gemm_wmma(
         input: *const c_void,         // device pointer [size_m, size_k]
         weights: *const c_void,       // device pointer [num_experts, size_n, size_k]
         sorted_token_ids: *const i32, // device pointer [size_m]

candle-kernels/src/moe/moe_wmma.cu

@@ -231,7 +231,7 @@ __global__ void moe_gemm_grouped_kernel(
         size_m, size_n, size_k \
     );\
 
-extern "C" void moe_gemm_wmma(
+extern "C" void candle_moe_gemm_wmma(
     const void* input,                // [size_m, size_k]
     const void* weights,              // [num_experts, size_n, size_k]
     const int32_t* sorted_token_ids,  // [size_m] (Device)

candle-nn/src/moe.rs

@@ -94,7 +94,7 @@ pub fn moe_gemm(
         use core::ffi::c_void;
 
         unsafe {
-            ffi::moe_gemm_wmma(
+            ffi::candle_moe_gemm_wmma(
                 input.device_ptr(input.stream()).0 as *const c_void, // [size_m, size_k]
                 weights.device_ptr(weights.stream()).0 as *const c_void, // [num_experts, size_n, size_k]
                 sorted_token_ids.device_ptr(sorted_token_ids.stream()).0 as *const i32,