fix: honor zero values in Dataset.reset_seeds

[alvinttang]

Dataset.reset_seeds updates each seed/size with value or self.value. Since 0 is falsy, passing a valid 0 is silently dropped and the previous value is kept.

Repro:

from dspy.datasets.dataset import Dataset
d = Dataset(train_seed=5, train_size=10)
d.reset_seeds(train_seed=0, train_size=0)
print(d.train_seed, d.train_size)  # prints "5 10", expected "0 0"

This affects anyone resetting to seed 0 (a common default) or requesting an empty split with size 0.

Fix: replace the or fallbacks with explicit is not None checks so 0 is applied while omitted (None) arguments still leave the existing value untouched. Added tests covering both the zero case and the omitted-argument behavior.

[greptile-apps]

Greptile Summary

This update correctly accepts explicit zero values for dataset seeds and split sizes, while preserving existing settings when arguments are omitted. Runtime checks confirmed that both evaluation-derived split seeds remain unchanged when eval_seed is not provided.

Confidence Score: 5/5

No blocking failure remains.

Focused runtime checks verified zero-value overrides, cached split invalidation, and preservation of both development and test seeds when evaluation settings are omitted.

T-Rex Logs

What T-Rex did

• Reproduced the historical regression by running the dataset-reset-seeds-regression verifier with the legacy implementation, which exited with code 1.
• Verified the current implementation by running the same verifier with --implementation current, which exited with code 0 and confirmed that six zero-valued arguments apply, cached splits resample to length zero, and omitting eval_seed preserves dev_seed and test_seed.
• No tracked source files were modified, and the authored verifier and its command captures were saved as artifacts.

_{Ran code and verified through T-Rex}

_{Reviews (2): Last reviewed commit: "Apply suggestion from @greptile-apps[bot..." | Re-trigger Greptile}

[isaacbmiller]

lgtm