v0.1.0: data pipeline, fate baselines and a fate-conditioned generative model

First tagged version of static2fate (S-BIAD2515, HeLa cells under a staurosporine dose series, AnnexinV fate at 6 h).

What it contains:

• Nuclear segmentation, 13-frame tracking and terminal AnnexinV registration for 30 wells (119 fields): 6,676 cells with a fate label.
• Single-frame fate baselines: within 19.5 nM, leave-one-field-out AUROC 0.51 at 0 min, 0.60 at 2 h, 0.78 at 4 h and 0.84 at 6 h. Handcrafted nuclear features match or beat frozen DINOv2.
• A frozen V-JEPA 2 video encoder does not read fate earlier than one frame of handcrafted features (difference -0.045, 95% CI -0.067 to -0.021 at 3.5 h).
• A conditional flow-matching model (9.4 M parameters) that generates both fates from the same noise; branches separate from 3 h, as in real dying cells.

Reproduce: uv sync then the scripts in scripts/. Tests: uv run pytest. Citation: see CITATION.cff.