C-bowman/inference-tools: 0.15.0 release

Introduced a new algorithm for tuning the simulation step-size in HamiltonianChain. The new approach stores the running mean and variance of the acceptance probability for all step-size values which are tested. Previously, this information was only stored for the current value of the step-size. Incorporating this extra information into the step-size selection improves the stability of the step-size value over the chain.

HamiltonianChain now has a simulation_steps argument which allows the number of steps used in the Hamiltonian dynamics simulation to be specified.

Various type-hinting improvements for HamiltonianChain.