linspace tests fail on master

[andreasnoack]

Probably because of JuliaLang/julia#18777. E.g.

julia> u=eps(zero(Float32));

julia> [Compat.linspace(-u,u,4);][3] == 0
false

[timholy]

Yes, that test was one of the (few) casualties of the rewrite. It was advertised, see

I was also able to restore some of the tests that I had ditched in the original implementation of this PR. The only tests that I weakened were for cases where step underflows.

[andreasnoack]

I guess we should just remove the tests here as well.