Fix various problems with architecture version handling

[serge-sans-paille]

Some architectures were not registered in the xsimd::all_architectures arch list.

The test for issorted was incorrect.

Some architectures add wrong version values but this went unnoticed because of the invalid check.

[serge-sans-paille]

@JohanMabille : I think I unveiled a serious issue there: due to bad ordering of the version, wasm was never selected as a the default arch when available, which in turns nullify most of its test (intel arch was tested instead!)

It's good that I caught this, but it's going to have some impact...

[JohanMabille]

[JohanMabille]

I think we don't have fallback implementation for all operations, therefore we cannot add new instruction sets and registers if we don't implement a minimal subset.

What I don't understand is that when Anutosh was implementing the support for WASM, the build was failing until he had implemented the minimal subset; so WASM was selected at some point. Also this explains why removing the workaround Thorsten implemented for arg / conj did not make the tests fail: at this point, we were in this configuration, where WASM was not selected anymore.

[anutosh491]

Yes I believe there was a minimalistic subset of operations. After I had implemented all of those, the build stopped failing.

[SylvainCorlay]

Thanks @serge-sans-paille! It is looking like it is going to be in working order with wasm.

[serge-sans-paille]

@SylvainCorlay yep, wasm fixed, I'm handling the final touches and this should be mergeable. @JohanMabille this deserves a minor release (patch + new features)

[serge-sans-paille]

\o/ hurray, that one was a tough one!