added maximum threshold

[NiekWielders]

Summary

added a maximum threshold to the positions_likelihood_from

Changes

only positions_likelihood_from

Testing

no extra tests added

Related Issues

[Jammy2211]

Thanks @NiekWielders! Before merging I wanted to double-check one specific concern that's bitten us in this area before: the threshold value flowing into min(threshold, maximum_threshold) can be either a numpy scalar or a JAX 0-d array depending on whether the Analysis was constructed with use_jax=True or use_jax=False, and Python's min() can behave differently on the two.

I traced and tested it:

1. threshold is produced by positions_threshold_from via factor * np.nanmax(positions_fits.max_separation_of_plane_positions). np.nanmax always returns a numpy scalar even when given a concrete JAX array, so threshold arrives here as a numpy scalar (or a Python float in the minimum_threshold early-return branch) regardless of the backend.
2. I empirically tested min(threshold, maximum_threshold) against numpy scalar, numpy 0-d array, JAX 0-d array (fp32 and fp64), and Python float — all combinations work and return a usable numeric value.
3. positions_likelihood_from is only called on a Result after the fit completes (i.e. outside any jax.jit boundary), so JAX tracer concretization isn't a concern.

Conclusion: the code is safe for both use_jax=True and use_jax=False. Merging as-is.

Minor style nit (not blocking, just for future reference): the existing minimum_threshold clamp lives inside positions_threshold_from using an if threshold < minimum_threshold: return minimum_threshold pattern, while this PR adds the symmetric maximum_threshold clamp inside positions_likelihood_from using min(). Asymmetric placement but functionally equivalent — happy to leave as-is.