Remove R2 extra allocations

[MohamedLaghdafHABIBOULLAH]

Three changes in order to make solve! of R2 non allocating.

• Define $\rho_k$ as local instead of global variable.
• Use the last version of ShiftedProximalOperatorsin order to use non allocating prox! of L0, L1, Rank and Nuclear norm.
• Add test_allocs.jl file based on @wrappedallocs macro.
  @dpo @MaxenceGollier

[MohamedLaghdafHABIBOULLAH]

Do you think it is beneficial to precise the type of statswithin R2 as follows ?

stats = GenericExecutionStats(reg_nlp.model, solver_specific = Dict{Symbol, Union{Float64, Vector{Float64}, Vector{Int64}}}())

[dpo]

@MohamedLaghdafHABIBOULLAH Yes it definitely would, but you don’t want to use Float64. Instead, use the element type of the NLPModel.

Also, just use Int instead of Int64 (Int = Int64 on 64-bit platforms).