Minibatching through`DataLoader` with `OptimisationOptimJL.AbstractOptimizer` is broken

[vboussange]

using Optimization, OptimizationOptimJL, OptimizationOptimisers

data = randn(2, 100)
idx_ranges = 1:size(data, 2)

loss(x, idx_rngs) = sum(abs2, data[:, idx_rngs] .- x)
objectivefun = OptimizationFunction(loss, Optimization.AutoForwardDiff())

x0 = [1.0, 1.0]
train_loader = DataLoader(idx_ranges; batchsize = 2, shuffle = true, partial=true)
optprob = OptimizationProblem(objectivefun, x0, train_loader)

# passes
opt = OptimizationOptimisers.Adam()
Optimization.solve(optprob,
                    opt,
                    maxiters=100)

# breaks
opt = OptimizationOptimJL.Adam()
Optimization.solve(optprob,
                    opt,
                    maxiters=100)