Add multi-start functionality to optimisers

[BradyPlanden]

Feature description

Add a multi-start method to the optimiser classes with a random initial value for each start. A few open questions

• Should multi-start be applied to all optimisers or only local optimisation algorithms?
• Can multi-start be multiprocessed to run the optimisations in a parallel pool?

Motivation

Multistarting local optimisers provides a defence against converging to a local optimal while missing a nearby global.

Possible implementation

A potential interface could look like:

optim = pybop.AdamW(cost, multistarts=10, starting_distribution=pybop.Uniform(0,1.0))
results = optim.run()

where results includes subdictionaries of the results for each multistart optimisation, and starting_distribution is the distribution to draw initial values from, if not provided, the parameter.prior distribution is used.

Additional context

No response