Polish up the evaluation scripts

[levmckinney]

• Fixes a bug that basically renders the eval script useless.
• Removes redundant options
• Switches metrics to being saved as json files instead of pt files
• Adds back in disaggregated batch metrics.

[codecov[bot]]

Codecov Report

Merging #92 (47358b1) into main (1308bdd) will increase coverage by 0.51%. The diff coverage is 87.05%.

@@            Coverage Diff             @@
##             main      #92      +/-   ##
==========================================
+ Coverage   79.57%   80.08%   +0.51%     
==========================================
  Files          32       32              
  Lines        1826     1888      +62     
==========================================
+ Hits         1453     1512      +59     
- Misses        373      376       +3     

Impacted Files	Coverage Δ
tests/scripts/test_integration.py	100.00% <ø> (ø)
tuned_lens/scripts/ingredients.py	85.63% <75.00%> (+0.08%)	:arrow_up:
tuned_lens/stats/logit_stats.py	93.10% <85.00%> (-2.65%)	:arrow_down:
tuned_lens/scripts/eval_loop.py	90.24% <87.71%> (+1.54%)	:arrow_up:
tuned_lens/scripts/train_loop.py	78.29% <100.00%> (+0.09%)	:arrow_up:

... and 1 file with indirect coverage changes