Fix land variable performance issue by eagerly loading data

[chengzhuzhang]

The pull request improves performance further based on the initial performance improvement work in #26. When testing 165 year historical runs, I found that the land variables were taking ~18 minutes each vs ~5 seconds for atmosphere variables. The issue was dask lazy evaluation - when area scaling arrays (total_land_area, north_land_area, south_land_area) remained as lazy dask arrays, the multiplication operation triggered loading all data from disk, causing the massive delay.

Solution: Eagerly load both area fields and computed data arrays into memory before performing operations. This ensures all operations work with numpy arrays instead of lazy dask arrays.

Changes:

• For TOTAL metric variables, call .load() on area fields (valid_area_per_gridcell, area, landfrac) after opening dataset
• Call .load() on annual average data_array after computation
• Reduces land variable processing from ~18 minutes to ~5-10 seconds

🤖 Generated with Claude Code

Select one: This pull request is...

• a bug fix: increment the patch version
• a small improvement: increment the minor version
• a new feature: increment the minor version
• an incompatible (non-backwards compatible) API change: increment the major version

Small Change

• To merge, I will use "Squash and merge". That is, this change should be a single commit.
• Logic: I have visually inspected the entire pull request myself.
• Pre-commit checks: All the pre-commits checks have passed.

[forsyth2]

This looks reasonable from visual inspection, and I see the GitHub Actions are passing.

Since we're in a heavy testing period for the Unified release anyway, we can check the integration tests results on main after merging / in the next rc.

[chengzhuzhang]

Thanks @forsyth2 .