Fix data loss in gridded_skill with binsize parameter

[ecomodeller]

Problem

Fixes #157

When using the binsize parameter in gridded_skill() without explicitly defining bin edges, the auto-generated bins don't cover the full spatial range of the data. This causes data points at the boundaries to be excluded from all bins, resulting in data loss.

Example

With track data spanning from latitude 31.4° to 65.62°, using binsize=0.25 without explicit bin edges causes approximately 4° of data (~400 km of satellite data) to be excluded.

Solution

Modified _add_spatial_grid_to_df() in src/modelskill/comparison/_utils.py to:

1. Calculate bin edges that cover the full data range (from min to max)
2. Align bins to multiples of binsize relative to the mean (maintains consistency)
3. Ensure no data points are excluded at the boundaries

Key Changes

• Replaced center-based bin calculation with min/max-based calculation
• Bins now start at or before the minimum data value
• Bins now extend to or beyond the maximum data value
• All 532 data points now included (previously 17 points were excluded)

Testing

• ✅ New test test_gridded_skill_binsize_no_data_loss verifies all data points are included
• ✅ Updated existing test expectation (5 bins instead of 4 - needed to cover full range)
• ✅ All gridded_skill tests pass
• ✅ All track comparison tests pass

Commits

1. Add failing test demonstrating the bug (17 points lost)
2. Implement fix and update test expectations