polykin 0.8.0

The main changes include:

• Extend module polykin.math:

  • add machine constants module polykin.math.machine and expose it at the top-level math namespace.
  • add optimization subpackage polykin.math.optimization with:
    • one-dimensional methods: fmin_brent, fmin_secant.
    • multidimensional methods: fmin_qnewton, fmin_nelder_mead.
    • shared global strategies/utilities: line_search, dogleg.
    • dedicated optimization result types: OptimumResult, VectorOptimumResult.

• Extend/refactor module polykin.math.fixpoint:

  • add new fixed-point solvers: fixpoint_damped, fixpoint_dem, fixpoint_steffensen.
  • update existing solvers (including Anderson and Wegstein implementations) for improved consistency with the new math infrastructure.

• Refactor module polykin.math.roots:

  • move q-Newton vector root solver implementation to polykin.math.roots.qnewton.
  • keep scalar/root result APIs while aligning internals with shared machine constants and optimization helpers.

• Improve numerical infrastructure in polykin.math.derivatives and polykin.math.special:

  • substantial updates to numerical differentiation routines.
  • internal cleanup of special-function implementations.

• Internal consistency updates across the codebase:

  • migrate numerical precision constants usage from polykin.utils.math to polykin.math.machine.
  • update dependent modules in kinetics, transport, thermo, properties, and copolymerization to use the new math utilities.

• Compatibility notes:

  • imports from polykin.math.roots.vector should be migrated to the current roots API.
  • code importing eps, huge, or tiny from polykin.utils.math should migrate to polykin.math.machine.