feat(externcopr/nvidaopticalflow): fix output and fix cu context issue on

diff thread,  nvof loader outptr share with mgb ptr, upgrade lastest nvof sdk

GitOrigin-RevId: 6e003c64282ef2f119402dc602f7a27beae4a776

src/serialization/impl/extern_c_opr.cpp

@@ -59,6 +59,8 @@ MGBDType dtype_cpp2c(DType dtype) {
             return MGB_DTYPE_FLOAT32;
         case DTypeEnum::Int32:
             return MGB_DTYPE_INT32;
+        case DTypeEnum::Int16:
+            return MGB_DTYPE_INT16;
         case DTypeEnum::Uint8:
             return MGB_DTYPE_UINT8;
 #if !MEGDNN_DISABLE_FLOAT16
@@ -75,6 +77,8 @@ DType dtype_c2cpp(MGBDType dtype) {
     switch (dtype) {
         case MGB_DTYPE_UINT8:
             return dtype::Uint8{};
+        case MGB_DTYPE_INT16:
+            return dtype::Int16{};
         case MGB_DTYPE_INT32:
             return dtype::Int32{};
         case MGB_DTYPE_FLOAT32:

src/serialization/include/megbrain/serialization/extern_c_opr.h

@@ -36,6 +36,7 @@ typedef enum MGBDType {
     MGB_DTYPE_UINT8,
     //! IEEE 754-based half-precision floating
     MGB_DTYPE_FLOAT16,
+    MGB_DTYPE_INT16,
 } MGBDType;
 
 typedef struct MGBTensorShape {