convertible types instead of same type

[JohanMabille]

Fixes #958

[wolfv]

Is this also going to work e.g. for uint16_t which doesn't yet exist in xsimd?

[JohanMabille]

yep. If uint16_tis the value_type of the LHS, the xassign_traits<E1, E2>::simd_size returns false, if it is the value_type of the RHS, xassign_traits<E1, E2>::forbid_simd will return false, preventing a simd assign.

[JohanMabille]

OK this requires a release of xsimd to work.

[wolfv]

quick other question: In this test for has_simd_apply we were using the return type to test
but with the complex types this obviously fails (for example when using imag() in the lambda)

https://github.com/QuantStack/xtensor/blob/bf7f425bd4c00d14f48d54fe73238798e5094ff4/include/xtensor/xassign.hpp#L270-L275

So we should test with the value_type of the arguments somehow ... did you do something like this somewhere else already?

[JohanMabille]

Yep, see here.
I'll fix the use_strided_loop struct.