PERF: FUTURE: Default default-constructor of SymmetricSecondRankTensor

[N-Dekker]

Follow-up to pull request #4469 commit 755cd10
"PERF: FUTURE: Default default-constructors of RGBPixel and RGBAPixel"

[blowekamp]

I assume this PR is similar to the RGBPixel improvement with the goal of improving the performance of the Image::Allocate method. Perhaps this is a case there the optimization should go into the Allocate method. In SimpleITK and when interfacing with other libraries and buffers, the Image container buffer is considered a raw buffer. Perhaps if certain conditions, c++ traits or itk PixelTraits then a special optimization in this method could occur where the buffer is initialized as some type of raw buffer, then case to the component type.

These initialized objects can be an insidious thing to track down as the behavior is not deterministic and varies with debug and release builds. So I have some concern with the proposed future changes.

[N-Dekker]

I assume this PR is similar to the RGBPixel improvement with the goal of improving the performance of the Image::Allocate method.

Indeed, I have indeed seen huge ("future") performance improvements of Image::Allocate from the similar RGBPixel PR (#4469), and I would expect similar ("future") performance improvements for SymmetricSecondRankTensor, with this commit. But it's not just that. It's also a matter of consistency between pixel types:

    TPixel pixel1; // May not be initialized. 
    TPixel pixel2{}; // Properly zero-initialized.
    TPixel pixel3 = TPixel{}; // Properly initialized.
    TPixel pixel4 = TPixel(); // Properly initialized.

As I mentioned at #4469 (comment)

Note that std::make_unique<TPixel>() and std::make_shared<TPixel>() will zero-initialize the created pixel(s) anyway, also with this PR.

Slightly off-topic, I think it would help if itk::Image would have an extra member function, AllocateZeroInitialized(), which would be equivalent to Allocate(bool initializePixels) with true as argument. It seems cleared to me to explicitly write image.AllocateZeroInitialized(), instead of image.Allocate(true). What do you think?