fix: change float32 to float64 in Sign operator to prevent precision loss

[digital-wizard48]

Summary

• The Sign class in qlib/data/ops.py was converting series data to np.float32 to handle bool type inputs
• This caused significant precision loss for large numbers (e.g., multiplying close price by volume)
• For example, Mul($close, $volume) could return 3989893056.0 instead of the correct 3989893260.0
• The fix changes the dtype from np.float32 to np.float64 which preserves the necessary precision

Fixes #61

---

This PR was auto-generated by Gittensor bot using Claude AI to fix a reported issue.