Build and implement new reward stack

[schampoux]

Currently we use a one-hot rewarding system for miners. The goal of this issue is to build, implement, and test new rewarding systems that are more dependent on the distribution of energies.

• Sorting energies in ascending order (most negative to 0). Assign rewards linearly such that the most negative energy gets a reward of 1 and energy 0 gets a reward of 0.
• Scale the rewards using min-max normalization, excluding the zero energy value. This can be done by removing 0 energy values, determine the min and max energies from the filtered set, and apply min-max normalization to scale the rewards between 0 and 1.
• Rewards can be scaled exponentially to emphasize lower energy values more strongly.
• Logarithmic scaling can be used to reduce the impact of high energy values and provide a smoother reward distribution.
• Softmax scaling to convert the energies into a probability distribution, which can be interpreted as scaled rewards.
• others.

[schampoux]

This plot was constructed using the reward distribution from all miners and jobs. Brian had mentioned steering the initiative toward scaling each simulations output to formulate rewards. This is to be done on a batch level per simulation.

[schampoux]

Update on the direction of this issue. https://discord.com/channels/799672011265015819/1234881153832321024/1246185898765320233

[steffencruz]

@mccrindlebrian can we close?