0.5 breaking changes

[longemen3000]

• Remove the sparse solver for Association, the cost of building sparse matrices is not worth it, specially with combining rules. if we implement that + an special, non-allocating rule for association with 1-site, 1-comp, will allow 0 allocations PCSAFT.

• Rename x_userlocations to x_userparams (@pw0908). now that we accept parameters, seems a better name indeed

• There is still the problem with what should be the output from tp_flash models. we need to return the volumes, and the phases. maybe as a named tuple?

• Multiparameter models (Unify all Single Empiric Helmoltz models under one struct #148)

[pw0908]

Some additional features to add (non-breaking):

• Calculate AADs for each experimental data set in parameter estimation.
• Export the parameters in a given model to csvs (making our lives easier when doing parameter estimation).
• Since we're messing around with the reader, add a feature to read in sigma profiles.

[longemen3000]

On the sigma profiles, if we don't include the meta field (a JSON file),then reading those seems straightforward. i could whip up a reader, but i would need some examples to see what are the possible universe of valid profiles to parse
On the parameter writing, we kind of already have a writer (ParamTable can write to disk) but given that now we can pass those parameters directly, i would say that we should deprecate that and adapt the current writer to take all the params in a model recursively.