Replace glMapBufferARB() with temp memory + glBufferSubDataARB().

Instead of letting OpenGL manage the memory mapping, allocate a temporary
block of memory upon VBO:map(). Upon VBO::unmap() the block is copied to
graphics memory using glBufferSubDataARB() and then released.

Hopefully this will fix some weird error with Windows XP + old drivers, where
VBOs can be created, but the mapped memory cannot be allocated resulting in
the sprite batch drawing nothing at all.

According to the interwebs this could actually faster than using OpenGL mapped
memory, because there is no unnecessary syncing involved.

src/modules/graphics/opengl/VertexBuffer.cpp

@@ -72,7 +72,7 @@ namespace opengl
 		delete [] buf;
 	}
 
-	void *VertexArray::map(GLenum/* access*/)
+	void *VertexArray::map()
 	{
 		return buf;
 	}
@@ -122,20 +122,21 @@ namespace opengl
 			unload(false);
 	}
 
-	void *VBO::map(GLenum access)
+	void *VBO::map()
 	{
-		// Don't map twice.
+		// mapping twice could result in memory leaks
 		if (mapped)
-			return mapped;
+			throw love::Exception("VBO is already mapped!");
 
-		mapped = glMapBufferARB(getTarget(), access);
+		mapped = malloc(getSize());
 
 		return mapped;
 	}
 
 	void VBO::unmap()
 	{
-		glUnmapBufferARB(getTarget());
+		glBufferSubDataARB(getTarget(), 0, getSize(), (void*)mapped);
+		free(mapped);
 		mapped = 0;
 	}
 
@@ -209,7 +210,7 @@ namespace opengl
 			GLint size;
 			glGetBufferParameterivARB(getTarget(), GL_BUFFER_SIZE, &size);
 
-			const char *src = static_cast<char *>(map(GL_READ_ONLY));
+			const char *src = static_cast<char *>(map());
 
 			if (src)
 			{