Add CompactPerformanceEvaluation objects and the option in evaluate! to construct them

[ablaom]

Partly addresses JuliaAI/MLJ.jl#1105, and prepares for JuliaAI/MLJTuning.jl#215.

This PR also drops the per_fold part of PerformanceEvaluation display. More often than not, the table does not fit with this in it. The display still includes the names of all accessible fields. This PR also adjusts the display of per_fold information to mitigate issue with overly wide table.

Previously:

PerformanceEvaluation object with these fields:
  model, measure, operation, measurement, per_fold,
  per_observation, fitted_params_per_fold,
  report_per_fold, train_test_rows, resampling, repeats
Extract:
┌─────────────┬──────────────┬─────────────┬─────────┬──────────────────────────────────────
│ measure     │ operation    │ measurement │ 1.96*SE │ per_fold                            ⋯
├─────────────┼──────────────┼─────────────┼─────────┼──────────────────────────────────────
│ BrierLoss() │ predict      │ 0.933       │ 0.184   │ [0.719, 1.16, 0.844, 1.16, 0.667, 0 ⋯
│ Accuracy()  │ predict_mode │ 0.0         │ 0.0     │ [0.0, 0.0, 0.0, 0.0, 0.0, 0.0]      ⋯
└─────────────┴──────────────┴─────────────┴─────────┴──────────────────────────────────────

After this PR:

PerformanceEvaluation object with these fields:
  model, measure, operation,
  measurement, per_fold, per_observation,
  fitted_params_per_fold, report_per_fold,
  train_test_rows, resampling, repeats
Extract:
┌───┬─────────────┬──────────────┬─────────────┐
│   │ measure     │ operation    │ measurement │
├───┼─────────────┼──────────────┼─────────────┤
│ A │ BrierLoss() │ predict      │ 0.933       │
│ B │ Accuracy()  │ predict_mode │ 0.0         │
└───┴─────────────┴──────────────┴─────────────┘
┌───┬──────────────────────────────────────────┬─────────┐
│   │ per_fold                                 │ 1.96*SE │
├───┼──────────────────────────────────────────┼─────────┤
│ A │ [0.719, 1.16, 0.844, 1.16, 0.667, 0.914] │ 0.184   │
│ B │ [0.0, 0.0, 0.0, 0.0, 0.0, 0.0]           │ 0.0     │
└───┴──────────────────────────────────────────┴─────────┘

To do:

• Check this does not interfere with logging (MLJFlow.jl tests ok)
• Run MLJ integration tests
• Changes are sufficient for planned MLJTuning.jl change (see Current performance evaluation objects, recently added to TunedModel histories, are too big MLJ.jl#1105)
• Check Stack not broken by addition of evaluate! arguments

[codecov]

Codecov Report

Attention: Patch coverage is 94.11765% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 88.13%. Comparing base (b6056dc) to head (8cb6f26).
Report is 6 commits behind head on dev.

❗ Current head 8cb6f26 differs from pull request most recent head c83cae9. Consider uploading reports for the commit c83cae9 to get more accurate results

Files	Patch %	Lines
src/machines.jl	50.00%	1 Missing ⚠️
src/resampling.jl	96.87%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##              dev     #973      +/-   ##
==========================================
+ Coverage   87.74%   88.13%   +0.38%     
==========================================
  Files          28       28              
  Lines        2457     2587     +130     
==========================================
+ Hits         2156     2280     +124     
- Misses        301      307       +6     

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

[ablaom]

For the second time codecov is getting cancelled. A timeout?

[OkonSamuel]

LGTM!!