Incorrect "invalid_mask" in "save_ply"

[Daaohame]

Hi, @MrNeRF @maturk

Thanks for adding save_ply (#234 #427).

However, there is a bug in save_ply: the invalid_mask calculation where np.isnan(opacities).any(axis=0) marks the entire mask as True, causing all data points to be invalid. This happens because opacities has shape (N,), and using axis=0 makes the any() function return a single boolean value instead of row-wise filtering.

Steps to Reproduce:

1. Run the code with an opacities array containing some NaN values.
2. Observe that invalid_mask becomes completely True, invalidating all rows. This means no points will be saved.

Suggested Fix:
Replace
np.isnan(opacities).any(axis=0) | np.isinf(opacities).any(axis=0)
with
np.isnan(opacities) | np.isinf(opacities).

Version: Commit 1a1e0cc

Thanks for the great repo! Let me know if I can help with a fix.