Type unstable when data is vector of vector

[avik-pal]

julia> interp = LinearInterpolation([rand(2) for _ in 1:2], rand(2))
LinearInterpolation with 2 points
┌────────────┬──────────┬──────────┐
│       time │       u1 │       u2 │
├────────────┼──────────┼──────────┤
│ 0.00804659 │ 0.662718 │ 0.208611 │
│   0.711882 │ 0.652273 │ 0.222546 │
└────────────┴──────────┴──────────┘

julia> interp(0.5)
2-element Vector{Float64}:
 0.6554169013438129
 0.21835102996775396

julia> @code_warntype interp([0.5])
MethodInstance for (::LinearInterpolation{Vector{Vector{Float64}}, Vector{Float64}, Vector{Float64}})(::Vector{Float64})
  from (interp::DataInterpolations.AbstractInterpolation)(t::AbstractVector) @ DataInterpolations ~/.julia/packages/DataInterpolations/NoIUa/src/DataInterpolations.jl:24
Arguments
  interp::LinearInterpolation{Vector{Vector{Float64}}, Vector{Float64}, Vector{Float64}}
  t::Vector{Float64}
Body::Vector{Any}
1 ─ %1 = Base.getproperty(interp, :u)::Vector{Vector{Float64}}
│   %2 = DataInterpolations.eltype(%1)::Core.Const(Vector{Float64})
│   %3 = DataInterpolations.eltype(t)::Core.Const(Float64)
│   %4 = DataInterpolations.promote_type(%2, %3)::Core.Const(Any)
│   %5 = DataInterpolations.similar(t, %4)::Vector{Any}
│   %6 = (interp)(%5, t)::Vector{Any}
└──      return %6

Type unstable when we supply A vector of vector as data. %2 = DataInterpolations.eltype(%1)::Core.Const(Vector{Float64}) seems incorrect; probably, it needs to be eltype(eltype(...)) or something like that.