Refactor conv op to use slightly safer std::vector

Inspired by similar change to recomb op in commit 60c5c50

src/operations.cc

@@ -164,10 +164,10 @@ namespace sharp {
    */
   VImage Convolve(VImage image, int const width, int const height,
     double const scale, double const offset,
-    std::unique_ptr<double[]> const &kernel_v
+    std::vector<double> const &kernel_v
   ) {
     VImage kernel = VImage::new_from_memory(
-      kernel_v.get(),
+      static_cast<void*>(const_cast<double*>(kernel_v.data())),
       width * height * sizeof(double),
       width,
       height,

src/operations.h

@@ -53,7 +53,7 @@ namespace sharp {
    * Convolution with a kernel.
    */
   VImage Convolve(VImage image, int const width, int const height,
-    double const scale, double const offset, std::unique_ptr<double[]> const &kernel_v);
+    double const scale, double const offset, std::vector<double> const &kernel_v);
 
   /*
    * Sharpen flat and jagged areas. Use sigma of -1.0 for fast sharpen.

src/pipeline.cc

@@ -1606,7 +1606,7 @@ Napi::Value pipeline(const Napi::CallbackInfo& info) {
     baton->convKernelScale = sharp::AttrAsDouble(kernel, "scale");
     baton->convKernelOffset = sharp::AttrAsDouble(kernel, "offset");
     size_t const kernelSize = static_cast<size_t>(baton->convKernelWidth * baton->convKernelHeight);
-    baton->convKernel = std::unique_ptr<double[]>(new double[kernelSize]);
+    baton->convKernel.resize(kernelSize);
     Napi::Array kdata = kernel.Get("kernel").As<Napi::Array>();
     for (unsigned int i = 0; i < kernelSize; i++) {
       baton->convKernel[i] = sharp::AttrAsDouble(kdata, i);

src/pipeline.h

@@ -197,7 +197,7 @@ struct PipelineBaton {
   std::unordered_map<std::string, std::string> withExif;
   bool withExifMerge;
   int timeoutSeconds;
-  std::unique_ptr<double[]> convKernel;
+  std::vector<double> convKernel;
   int convKernelWidth;
   int convKernelHeight;
   double convKernelScale;