[CODE IMPROVEMENT] Add notice when the number of samples is cut in the Validation Insights tab

[pascal-pfeiffer]

🔧 Proposed code refactoring

We limit the samples to 900 in the Validation Insights tab to prevent lagging issues for very large datasets.

    elif len(df) > 900:
        df = df.sample(n=900, random_state=42).reset_index(drop=True)

We should show a notice when the subsampling on the Validation Insights tab is happening and point towards the "Download predictions" option.

Motivation

While users may still download all predictions it may be puzzling to see a different number here than in the training logs.

[us8945]

@pascal-pfeiffer , when we display Validation insights, the data is already sampled (if the number of validation records above 900).
With the change, when we display the message to the user, should we indicate exact number of the validation records available, or the message can just state that the number of validation records is above 900?