Allow free compositional vectors in optimal thermobarometry problems

[bobmyhill]

This pull request introduces support for compositional degrees of freedom in thermobarometric inversion, allowing mineral phases to vary their site occupancies during optimization for a better fit. It also refactors how site occupancies are represented and printed, adds new options to the site_formula method for Solution instances, and updates example scripts and tests to demonstrate and verify these new capabilities.

Major changes:

• Added support for free compositional vectors in the estimate_conditions function, enabling optimization over both P-T and compositional parameters. Phases can now have a free_compositional_vectors attribute, which is used to adjust their compositions during fitting. The optimization routine and parameter counting are updated accordingly.
• Changed the site_occupancies property in solution.py to return a flat numpy array instead of a list of dicts. The site_formula method now accepts precision, absent species printing, and custom site occupancies, making it more flexible for reporting and analysis.
• Updated example_optimal_thermobarometry.py to demonstrate how to use free compositional vectors for minerals like biotite and ilmenite, and improved output formatting for site occupancies and compositions, including more readable site names.
• Added new tests for the updated site_formula method, covering custom precision, absent species, and custom site occupancies. Also added a test for the new compositional vector optimization in thermobarometry.