This pull request:

• Adds a new constant logish_eps in burnman/constants.py as a mutable list holding a small epsilon value for numerical stability, and updates logish and inverseish functions in burnman/classes/solutionmodel.py to use this value
• Modifies the equilibrate function in burnman/tools/equilibration.py to retry a nonlinear solve with a relaxed logish_eps if the solver hits the maximum iteration limit. The process chains up to three solves and combines their results.
• Refactors the setup in equilibrate to use named functions for the system F, J and constraints.
• Adds a new test that uses the new fallback routine.