[R-package] allow for small numerical differences in Booster test

[jameslamb]

See #4680 for background on this PR.

On the submission of v3.3.0 to CRAN, one test in the R package failed in one CRAN test setup (r-devel-windows-x86_64-gcc10-UCRT).

── 1. Failure (test_lgb.Booster.R:469:5): Booster$eval() should work on a Datase
eval_in_mem not identical to eval_from_file.
Objects equal but not identical

This PR proposes using testthat::expect_equal() instead of testthat::expect_identical() on Windows in the relevant test, to reduce the risk of a rejected submission caused by small numerical precision differences.

Notes for Reviewers

As noted in #4680, LightGBM's CI does not currently have a job replicating CRAN's UCRT tests, and I'm not aware of an easy way to replicate that job. Also, confusingly, CRAN re-ran the package's tests in that environment and the second time, they all passed: #4680 (comment).

With this PR, I'm just trying to be overly-cautious in the hope that it reduces the risk of {lightgbm} being removed from CRAN. I think that most users of the R package would prefer very small numeric precision issues on Windows to the package being entirely unavailable.

[StrikerRUS]

Thanks a lot for the detailed description! It makes sense to me.

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.