A few optimizations

[schauveau]

Generally speaking, libopenshot is well written and quite efficient but I noticed a few things that could be optimized:

• ImageReader is creating a Magick::Image that is converted into a new QImage for each frame (via Frame::AddMagickImage). That conversion is not very efficient and could be avoided by reusing the same QImage.
• QImage::bits() and QImage::scanline() provide a non-const pointer to the image data. If the QImage is shared (because of a shallow copy), this causes an automatic duplication of the data. Consequently, if the data is not going to be modified then it could be beneficial to use QImage::constBits() or QImage::constScanline() instead. Potential gains are in Frame::GetPixels(), Frame::AddImage(std::shared_ptr<QImage> new_image, bool only_odd_lines), Frame::GetPixels(int row). and maybe a few other places.
• In several places, the pixels of a QImage are processed with a loop of the form for (n = 0; n < image->columns() * image->rows(); n += 1). This is highly inefficient because QImage::rows() and QImage::columns() are functions that cannot be inlined by the compiler. Consequently, the compiler cannot assume that their values are constant and must perform the 2 calls at each iteration. Also, the code cannot be vectorized. This is found in Timeline::add_layer and in most of the effect classes. A similar problem can be found in Frame::AddMagickImage in the loop that converts all pixels of the Magick::Image to RGBA.

[ferdnyc]

• QImage::bits() and QImage::scanline() provide a non-const pointer to the image data. If the QImage is shared (because of a shallow copy), this causes an automatic duplication of the data. Consequently, if the data is not going to be modified then it could be beneficial to use QImage::constBits() or QImage::constScanline() instead. Potential gains are in Frame::GetPixels(), Frame::AddImage(std::shared_ptr<QImage> new_image, bool only_odd_lines), Frame::GetPixels(int row). and maybe a few other places.

On this point: As I noted in #375, QImage::bits() and QImage::scanLine() are overloaded functions, and in const context they're equivalent to constBits() and constScanLine() automatically. We merged the change to use the const...() forms explicitly, for the sake of code readability and clarity, but it was purely code-cosmetic — the const context of the previous calls meant that they were already avoiding any unnecessary copying.