CodeUnits <: DenseVector, but does not fulfill its only criteria

[jakobnissen]

The only (definitional) criteria of DenseArray is that it's stored contiguously in memory.
We also have Codeunits <: DenseVector. However, there is no requirement that the string backing the codeunits are stored densely in memory. For example, InlineString.jl provide string types which are 8-byte primitives.

I'm not sure what to do about this. It might be breaking to remove Codeunits <: DenseVector, but it's also silly to have it subtype DenseVector when the only thing that matters about DenseVector is not upheld.

[jakobnissen]

See discssion in #26085

[mbauman]

julia/base/strings/basic.jl

Lines 793 to 796 in 0e28cf6

	struct CodeUnits{T,S<:AbstractString} <: DenseVector{T}
	s::S
	CodeUnits(s::S) where {S<:AbstractString} = new{codeunit(s),S}(s)
	end

Ooof, that's tricky. I guess we could try the simple thing — change it to AbstractVector — and check both of nanosoldier's reports. If we need to preserve CodeUnits{UInt8, String} <: DenseVector{UInt8} the realm of possible fixes (and their associated breakages) becomes much more complicated and the tradeoffs will be hard to evaluate.