feat(distance): add na_comp option aligned with baseR dist behavior for NA handling

[SpatLyu]

Previously, when na_rm = true, the distance calculation would simply skip NA/NaN dimensions
and sum over the remaining valid dimensions. This creates a systematic bias: pairs with more missing
values would artificially appear "closer" than they truly are, simply because fewer dimensions
contributed to the total distance.

This behavior is inconsistent with base R's dist() function, which applies proportional scaling when computing (Euclidean, Manhattan, Canberra or Minkowski) distances with missing values. This inconsistency can lead to:

• Unexpected results for users migrating from base R
• Biased downstream analyses (e.g., Information Imbalance, KNN-based methods)
• Difficulty in reproducing results across different R packages

Changes

• Added na_comp parameter to Euclidean and Manhattan distance functions in C++ backend
• When na_comp = true and na_rm = true, the computed distance sum is scaled up proportionally:
  adjusted_distance = raw_distance * (total_dims / valid_dims)