Make `suggested_init`, `suggested_bounds` and `suggested_fixed` property attributes

[matthewfeickert]

Similarly to this, suggested_init, suggested_bounds and suggested_fixed currently take no arguments either. One other consideration is that maybe some of these parts of the API are meant to eventually support kwargs, in which case switching to a property would not be useful.

Originally posted by @alexander-held in #1700 (comment)

This is a good idea and should be done, but this would cause more breaks in pyhf v0.7.0 then I think we want

$ git grep --name-only "suggested_init()" | wc -l
42

So I think these should hold off until v0.8.0

[nikoladze]

I personally would find it useful if these supported passing "overrides" via keyword arguments, e.g. suggested_init(mu=1) would give me the set of parameters with "mu" set to 1 and the rest to the suggested initial values. I regularly copy over small helper functions along these lines:

from functools import partial

def with_override(fn, **kwargs):
    pars = list(fn())
    for k, v in kwargs.items():
        pars[model.config.par_slice(k)] = [v]
    return pars

init = partial(with_override, model.config.suggested_init)
fixed = partial(with_override, model.config.suggested_fixed)
bounds = partial(with_override, model.config.suggested_bounds)

Some thinking needs to be done how to handle parameter sets with slices spanning more than one value (e.g. staterror), e.g. one could require to pass all values for the parameter set as a list then.

[alexander-held]

I like that idea a lot! @MoAly98 proposed a fairly related feature for cabinetry in scikit-hep/cabinetry#524 to address this pretty common use case of only wanting to change a small number of parameters. Having this centrally in pyhf could perhaps largely address this for cabinetry users as well.