random_seed not propagated to randomise_axial_position (dormant reproducibility gap)

Follow-up from the random-seed propagation work in #134.

What
CoordinatesField.randomise_axial_position() (src/vlab4mic/generate/coordinates_field.py) builds its RNG as np.random.default_rng(seed=random_seed), but its only caller, _gen_abs_from_rel_positions() (line ~501), invokes it with no argument:

if self.axial_offset is not None:
    self.randomise_axial_position()   # random_seed defaults to None

So when axial offsets are in use, the axial z-placement draws from a fresh non-deterministic Generator and ignores the seed (a default_rng(None) does not read the global np.random.seed set in experiments.py). All other stochastic inputs (positions via the global seed, orientations and rotations via the threaded random_seed) are reproducible after #134.

Impact: dormant. randomise_axial_position only runs when self.axial_offset is not None. No current article-figure script sets axial_offset (the NPC scripts use axial_precision / axial_resolution_nm, which are unrelated PSF params), so the manuscript figures are unaffected and reproducible. This only bites a user who explicitly configures axial_offset.

Suggested fix
_gen_abs_from_rel_positions() is a no-arg helper called from many places, so threading the seed through the call site is awkward. Cleaner: persist the seed on the field instance (e.g. set self.random_seed in set_molecules_params / create_minimal_field) and have randomise_axial_position default to self.random_seed. Then drop the now-redundant random_seed params that are accepted but unused in the global-np.random position helpers (generate_random_positions), or wire them similarly, for a consistent seeding story.

Priority: low (latent; does not affect current figures).

Start in src/vlab4mic/generate/coordinates_field.py, reading CoordinatesField.randomise_axial_position() and its call from _gen_abs_from_rel_positions() around line 501. Trace how set_molecules_params or create_minimal_field handles the seed, then verify that configurations using axial_offset produce repeatable axial placement with the same seed without breaking the existing position and orientation behavior.