Add auto_mask parameter to Model class

[FBumann]

Disclaimer

Im not sure how much value this adds, as i didnt use masking that much before. Im happy to discuss the ups and downsides of such a feature. I would have raised an issue, but i thought adding a benchmark and proof of concept would be more valuable.

Summary

Adds auto_mask parameter to Model that automatically masks variables and constraints where input data contains NaN values.

# Before: manual masking required
m = Model()
mask = gen_capacity.notnull()
x = m.add_variables(lower=0, upper=gen_capacity, mask=mask, name="x")

# After: automatic
m = Model(auto_mask=True)
x = m.add_variables(lower=0, upper=gen_capacity, name="x")  # NaN auto-masked

Benefits

• Convenience: No manual mask creation/tracking
• Safety: Can't forget to mask NaN values
• Performance: Uses optimized numpy internals (np.where instead of xarray where = 38x faster)

Auto-mask conditions

Variables masked out where:

• lower is NaN, OR
• upper is NaN

Constraints masked out where:

• All variable references are invalid (null expression), OR
• rhs is NaN

Performance (30% NaN, includes external mask creation)

Variables

Potential	Active	No Mask	Manual	Auto
10K	7K	14ms	2ms	2ms
100K	70K	2ms	2ms	2ms
500K	350K	3ms	3ms	3ms
2M	1.4M	6ms	6ms	5ms

Constraints

Count	Terms	No Mask	Manual	Auto
5K	500	17ms	16ms	14ms
10K	1000	44ms	45ms	30ms
20K	1000	109ms	61ms	46ms
50K	500	171ms	101ms	61ms

Auto-mask is equal or faster than manual masking due to optimized numpy operations.

Checklist

• Code documented
• Tests added (62 pass)
• Release notes added
• MIT license consent