Multi-backend Parallel execution mode of the OptimizationSolver

[GaboFGuerra]

User Story

As a user of the OptimizationSolver I want the Optimization solver with many parallel configurations running simultaneously to get higher solution probability and a lower time to solution on both Loihi2 and CPU backends.

Conditions of satisfaction

• User can explicitly invoke the number of parallel copies of the solver network in SolverConfig.
• At runtime, the network state is read from the network with the best cost.
• The user can give the solver a set of hyperparameters in SolverConfig which parameterize each copy of the network solver.

Acceptance Tests

For Loihi2 and CPU backends

• Number of parallel copies of the network in SolverConfig results in an equivalent number of SNN + CostIntegrator copies on the SolverProcess and SolverModel.
• ReadGate's RefPorts for solution readout and spike counters for cost reading are equal the number of copies of the network.
• Solver network and solution readout processes are correctly connected.
• The input set of hyperparameters are correctly propagated and distributed into the solver network copies.
• At runtime, the network state is read from the network with the best cost.

Tasks

• Solve issue Parallel execution mode of the CPU OptimizationSolver #170
• Solve issue Loihi 2 backend support for parallel OptimizationSolver  #171

Lava version:

• 0.3.0 (feature release)

I'm submitting a ...

• feature request

Other information:

insert the output from lava debug here