For exa backend, don't replace nothing by +/- Inf * ones(length(...)) but just by +/- Inf

EDIT: do it. seems like sth like

x[1]^2(t) + x[2]^2(t) <= 1

errors (sometimes...) as lb = [-Inf] while ub = 1...

• add the test from pr-adding-tensors #182

EDIT:

• don't do it for now (= scalar constraint): scalar constraint implies scalar bound; passing a vector here would be interpreted as a vector for the constraint generator!
• use $e1/3[1] (will scale to vector case with for loop on dimension length
• check all $pref.constraint(..., lcon=..., ucon=...) where lcon and ucon must be vectorised along generator for now, not dimension; these should be cases where length(e1) == length(e3) == 1

FTR: ExaModels.constraint lcon / ucon accept Number, AbstractArray, or Generator