LBFGS should have its own meta

[abelsiqueira]

For instance, nlp.meta.nnzh should be 0.

[dpo]

Actually, it’s more likely to be $n * (n+1) / 2$. I think it would be better to implement getters for meta instead of accessing fields directly.

[tmigot]

Actually, such getters already exist https://jso.dev/NLPModels.jl/dev/reference/#NLPModels.get_nnzh-Tuple{NLSMeta} for instance, but are not really used (yet?) and are sometimes buggy, for instance JuliaSmoothOptimizers/NLPModels.jl#415 ).