`randomMPS` fails to create states with `linkdim>1` on heterogenous lattice when qns are conserved

[MasonProtter]

Basically, if I have a lattice where alternating sites have different sitetypes, randomMPS(s, f; linkdims=k) for k>1 will end up erroring. Here is a MWE:

using ITensors

function ITensors.space(::SiteType"Sf=1/2";
                        conserve_qns=false)
    if conserve_qns
        return [QN("Sz", 1)=>1, QN("Sz",-1)=>1]
    end
    return 2
end

function ITensors.space(::SiteType"Sb=1/2";
                        conserve_qns=false)
    if conserve_qns
        return [QN("Sz", 2)=>1, QN("Sz",-2)=>1]
    end
    return 2
end

ITensors.state(::StateName"Up", ::SiteType"Sf=1/2") = [1.0, 0.0]
ITensors.state(::StateName"Dn", ::SiteType"Sf=1/2") = [0.0, 1.0]

ITensors.state(::StateName"Up", ::SiteType"Sb=1/2") = [1.0, 0.0]
ITensors.state(::StateName"Dn", ::SiteType"Sb=1/2") = [0.0, 1.0]

let 
    N = 4
    s = siteinds(n -> isodd(n) ? "Sf=1/2" : "Sb=1/2", N; conserve_qns=true)               
    ψ = randomMPS(s, n -> isodd(n) ? "Up" : "Dn"; linkdims=4)
end

#+RESULTS:
MPS center bond dim less than requested

Stacktrace:
 [1] error(s::String)
   @ Base ./error.jl:33
 [2] randomizeMPS!(M::MPS, sites::Vector{Index{Vector{Pair{QN, Int64}}}}, linkdim::Int64)
   @ ITensors ~/.julia/packages/ITensors/XnM47/src/mps/mps.jl:126
 [3] randomMPS(ElType::Type, sites::Vector{Index{Vector{Pair{QN, Int64}}}}, state::Function; linkdims::Int64)
   @ ITensors ~/.julia/packages/ITensors/XnM47/src/mps/mps.jl:213
 [4] #randomMPS#829
   @ ~/.julia/packages/ITensors/XnM47/src/mps/mps.jl:207 [inlined]
 [5] top-level scope
   @ In[7]:28
 [6] eval
   @ ./boot.jl:373 [inlined]
 [7] include_string(mapexpr::typeof(REPL.softscope), mod::Module, code::String, filename::String)
   @ Base ./loading.jl:1196

Important to this example is that Sb=1/2 sites carry a flux of two times greater than Sf=1/2 sites. If they carry the same flux, then the problem disappears.

[mtfishman]

This isn't particular to heterogeneous lattices, it can also happen with homogeneous lattices with certain starting states and desired bond dimensions. I think we should just make this a warning instead of an error while we work out a more substantial fix.