Clarify limit_{train,val,test}_batches behaviour (Lightning-AI#2630)

Author: jonashaag

docs/source/fast_training.rst

@@ -74,6 +74,8 @@ If you don't want to check 100% of the training/validation/test set (for debuggi
         limit_test_batches=0.3
     )
 
+If you also pass ``shuffle=True``, a different random subset of your dataset will be used for each epoch; otherwise the same subset will be used for all epochs.
+
 .. note:: ``limit_train_batches``, ``limit_val_batches`` and ``limit_test_batches`` will be overwritten by ``overfit_batches`` if ``overfit_batches`` > 0. ``limit_val_batches`` will be ignored if ``fast_dev_run=True``.
 
 .. note:: If you set ``limit_val_batches=0``, validation will be disabled.