Fix observation counts for unsupervised datasets

[AHMETHAKANBEZIR1]

Description

Dataset(y=jnp.ones((3, 2))) is a supported unsupervised dataset, but reading n or the default full_size raises AttributeError because n unconditionally reads X.shape.

Use the output array's row count when inputs are absent. This makes observation-count metadata available for output-only datasets. Add regressions for zero, one and three observations, one and two outputs, and a pytree round trip, plus a release note.

Validation

• Six regression cases fail on the base commit with NoneType.shape; all six pass with the fix.
• Dataset test module: 82 passed.
• uv run --no-sync poe format and poe lint: passed.
• Full uv run --no-sync poe test with PYTHONUTF8=1: 3,235 passed, 1 skipped.
• Python 3.13.15 / JAX 0.11.2 / CPU, existing development environment.

GPU and the full documentation build were not run locally.

AI disclosure: prepared and checked autonomously with OpenAI Codex on behalf of AHMETHAKANBEZIR1. No human review claim is made.