Add zip_iterator implementation

[upsj]

This PR adds a generic zip_iterator implementation as a replacement for IteratorFactory, which is currently limited to only two types with a fixed comparison operator. Most of the code is boilerplate, a few important things are:

• zip_iterator_reference is basically a std::tuple of references, so any assignment to it goes directly to the underlying values. This has the same caveats and downsides as the std::vector<bool> bit reference type and, coincidentally, accessors, i.e.
  • auto a = *it returns a value for normal iterators, but a reference-like type in our case, so subsequent assignment to a writes to the underlying memory!
  • auto& a = *it leads to a dangling reference (unless you make it const, which extends the lifetime AFAIK)
  • std::swap(a, b) for two such references leads to one of the values written to both. This is why we provide a swap implementation ourselves. The suggested way to use it in generic code is using std::swap; swap(a, b) to enable the namespace-local function to be found first via ADL.
• The comparison and difference operators are implemented such that in debug builds, they assert that the different iterators in the tuple provide the same result to comparisons (in release we just use the first iterator). This behavior is checked using GTest death tests.

Luckily, the C++ standard library already deals correctly with these issues by always writing to value_type directly or using only functions like swap for actual operations on entries.

A nice side-effect is that it gets rid of the -Wpsabi warnings related to the std::complex ABI, since everything now happens inside system headers, which AFAIK don't report warnings.

I guess @thoasm is probably most experienced with these kinds of constructions, but I can also provide some pointers to whoever else wants to review it :)

motivated by a comment by @pratikvn, implemented as a nice excercise in C++ edge cases :)

• Add some comments

[codecov]

Codecov Report

Merging #966 (3a78321) into develop (b34ae57) will decrease coverage by 0.00%.
The diff coverage is 99.20%.

@@             Coverage Diff             @@
##           develop     #966      +/-   ##
===========================================
- Coverage    92.32%   92.32%   -0.01%     
===========================================
  Files          480      480              
  Lines        39739    39750      +11     
===========================================
+ Hits         36691    36700       +9     
- Misses        3048     3050       +2     

Impacted Files	Coverage Δ
omp/matrix/csr_kernels.cpp	81.15% <ø> (ø)
omp/matrix/fbcsr_kernels.cpp	62.22% <ø> (ø)
omp/matrix/sparsity_csr_kernels.cpp	92.20% <ø> (ø)
reference/matrix/sparsity_csr_kernels.cpp	92.00% <ø> (ø)
reference/matrix/csr_kernels.cpp	89.91% <66.66%> (-0.22%)	⬇️
core/base/iterator_factory.hpp	100.00% <100.00%> (ø)
core/test/base/iterator_factory.cpp	100.00% <100.00%> (ø)
core/test/utils/unsort_matrix.hpp	77.77% <100.00%> (-0.61%)	⬇️
reference/matrix/fbcsr_kernels.cpp	97.18% <100.00%> (ø)
omp/reorder/rcm_kernels.cpp	97.53% <0.00%> (-0.61%)	⬇️
... and 1 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b34ae57...3a78321. Read the comment docs.

[thoasm]

Excellent implementation!
I mainly have a question regarding the assert functions for the compare operators.

[thoasm]

LGTM!

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
14 Code Smells

95.3% Coverage
4.5% Duplication

[tcojean]

LGTM! Nice work