[WTF] Fix performance when creating Vectors from different constness

https://bugs.webkit.org/show_bug.cgi?id=299008

Reviewed by Chris Dumez.

For example, when you try to create a Vector<uint8_t> from a span<const uint8_t>, templates will consider both types
different and even when you pass true to the template, the second VectorCopier::uninitializedCopy function will be used
and will defer to the slow version.

My use case was in MSE, creating a SharedBuffer implies creating a FragmentedSharedBuffer and then a DataSegment that
implicitly creates a Vector<uint8_t> from a span<const uint8_t>. If you're copying 2.5MB, time is reduced to 33% of the
initial one.

* Source/WTF/wtf/Vector.h:

Canonical link: https://commits.webkit.org/300464@main

Author: calvaris

Source/WTF/wtf/Vector.h

@@ -185,13 +185,14 @@ struct VectorCopier<false, T>
 template<typename T>
 struct VectorCopier<true, T>
 {
-    static void uninitializedCopy(const T* src, const T* srcEnd, T* dst)
+    static void uninitializedCopy(const T* src, const T* srcEnd, std::remove_const_t<T>* dst)
     {
-        memcpy(static_cast<void*>(dst), static_cast<void*>(const_cast<T*>(src)), reinterpret_cast<const char*>(srcEnd) - reinterpret_cast<const char*>(src));
+        memcpy(static_cast<void*>(dst), static_cast<void*>(const_cast<std::remove_const_t<T>*>(src)), reinterpret_cast<const char*>(srcEnd) - reinterpret_cast<const char*>(src));
     }
     template<typename U>
     static void uninitializedCopy(const T* src, const T* srcEnd, U* dst)
     {
+        static_assert(!std::is_same_v<std::remove_const_t<T>, std::remove_const_t<U>>, "We should be using the faster overload for this, which uses memcpy()");
         VectorCopier<false, T>::uninitializedCopy(src, srcEnd, dst);
     }
 };