chore: Remove caching logic from `CrossValidationReport` and `ComparisonReport`

[auguste-probabl]

What would you like to say?

Since these are essentially lists of EstimatorReports, they already benefit from the caching at the EstimatorReport level, so the added complexity adds very little performance benefits.

[thomass-dev]

Please note that the cache is currenlty used to retrieve permutation importance for both EstimatorReport and CrossValidationReport reports.

[cakedev0]

I 100% agree with this issue.

---

@thomass-dev

Please note that the cache is currenlty used to retrieve permutation importance for both EstimatorReport and CrossValidationReport reports.

In skore-hub-project right? The logic could be adapted for CrossValidationReport: we could inspect the cache of children EstimatorReports, and then call cv_report.inspection.permutation_importance() if we see that this will just read from cache.

And in the long term, the plan is to stop inspecting a private attribute of reports in skore-hub-project, right? (what you proposed in #2200 should go in this direction).

[thomass-dev]

I also agree with the issue. My previous message was just a point to consider and address during the development.

In skore-hub-project right?

Yes, and we can easily change skore-hub-project without breaking anything: at worst, the permutation is not sent to the hub.

the plan is to stop inspecting a private attribute of reports in skore-hub-project, right?

In the long term, yes, totally. But at short term, the cache could also become a public object.