forge.jit

Fast custom loops and reused results: ``@fg.jit`` and ``@fg.cached``.

``@fg.jit`` compiles a plain function over arrays and numbers to machine code, so a loop in ``compute()`` runs at native speed. ``@fg.cached`` remembers a function's result for one sweep, so a setting shared by many combinations is computed once. Nothing here imports ``btpro``, and ``numba`` is imported only when a function is compiled, so importing ``forge`` stays fast.