Return a data.frame from loo_compare() for subsampled loo objects

[florence-bockting]

Closes #392

What

loo_compare() returns a matrix for subsampled loo objects, but a data.frame
for all other loo objects. This PR aligns the two.

loo_compare.psis_loo_ss_list() now builds the same structure as
loo_compare() in R/loo_compare.R.

Output

Before:

       elpd_diff se_diff subsampling_se_diff
model2  0.0       0.0     0.0
model1 -16.5     22.5     0.4

After:

  model elpd_diff se_diff subsampling_se_diff
 model2       0.0     0.0                 0.0
 model1     -16.5    22.5                 0.4

Breaking change

Code that indexes the result by position breaks. comp[, 1] now returns the
model names. Use comp$elpd_diff. Code that reads rownames(comp) also
breaks. Use comp$model.

Checks

• devtools::test(filter = "loo_subsampling") passes.
• Snapshots in tests/testthat/_snaps/loo_subsampling_cases.md updated.
• vignettes/loo2-large-data.Rmd output blocks updated.
• reverse dependency checks

AI assistance

I used AI assistance while working on this PR. I have reviewed all changes.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.87%. Comparing base (8635a88) to head (8a8f767).

Additional details and impacted files

@@                 Coverage Diff                 @@
##           fix-subsampling     #393      +/-   ##
===================================================
+ Coverage            92.86%   92.87%   +0.01%     
===================================================
  Files                   31       31              
  Lines                 3041     3046       +5     
===================================================
+ Hits                  2824     2829       +5     
  Misses                 217      217              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 8a8f767 is merged into master:

• ✔️loo_function: 1.94s -> 1.94s [-0.77%, +0.47%]
• ✔️loo_matrix: 1.76s -> 1.75s [-1.53%, +0.87%]
  Further explanation regarding interpretation and methodology can be found in the documentation.