`isapprox` returns wrong results with `nans=true`

[gerlero]

using StaticArrays # v1.5.16

v = [1.0, NaN]
s = @SVector [1.0, NaN]

# These should be false
isapprox(v, v) # OK
isapprox(v, s) # OK
isapprox(s, v) # OK
isapprox(s, s) # OK

# But these should return true
isapprox(v, v, nans=true) # OK (no StaticArrays)
isapprox(v, s, nans=true) # wrong - returns false
isapprox(s, v, nans=true) # wrong - returns false
isapprox(s, s, nans=true) # wrong - returns false

[gerlero]

I could further trace the issue to this difference in the result of LinearAlgebra.norm with a StaticArray argument:

using StaticArrays
using LinearAlgebra

norm([0.0, NaN]) # NaN
norm([NaN, 0.0]) # NaN

norm(SA[0.0, NaN]) # 0.0
norm(SA[NaN, 0.0]) # NaN

[rmkn85]

Same with Inf

norm([1.0, Inf])    # Inf
norm(SA[1.0, Inf])  # NaN

[rmkn85]

The behaviour changed between v1.5.12 and v1.5.13 !!!

[rmkn85]

I think these are related:
#913
#975

[gerlero]

@rmkn85 thanks! I'll open a PR that should solve both cases (NaN and Inf)