Add profile-curve scan helpers and vectorize Neyman belts

[JaLuka98]

This pull request introduces several new features and improvements to the Neyman construction and profile likelihood workflow, focusing on performance and usability. The main updates include new utility functions for generating and inverting Neyman belts using precomputed grids, enhancements to the simulation API, and more efficient computation of profile likelihood curves.

Neyman construction and profile likelihood enhancements:

• Added build_belts_from_grid to neyman.py, allowing multiple Neyman belts to be constructed efficiently from a single toy ensemble and grid scan.
• Added invert_belt_from_curve, which enables inverting a Neyman belt using a precomputed observed q curve, avoiding repeated likelihood fits.
• Added profile_curve_from_likelihood and profile_curve_from_grid to stats.py for efficient evaluation of the profile likelihood (NLL) on a parameter grid without repeated minimization.

API and usability improvements:

• Updated the public API in __init__.py to export the new functions: build_belts_from_grid, invert_belt_from_curve, profile_curve_from_likelihood, and profile_curve_from_grid. [1] [2]
• Refactored GaussianModel.simulate to support generating multiple pseudo-experiments at once via a new size argument, and exposed the covariance and inverse covariance matrices as properties for easier access.

[JaLuka98]

Tests are failing because they are written for the current main version with an mle fit, now the grid leads to a boundary condition, this will be fixed later