The L4A crop type processor was developped to check the reliability of the crop type declarations. So the user imports all the parcels that she/he wants to check. Then, the processor will use part of the parcels to calibrate the classification model using the preprocessed EO data (by default 25% of the parcels for the main crop types) and will apply the classification model to all the parcels to predict the crop type for each parcel based only on the EO data. So it was developped to ingest a large number of parcels with crop declarations and check the reliability of these crop declarations (CAP context). Still, you can use it as you want, for example to extract markers by parcel and then make your own classification process. Or to build a classification model based on the parcels with crop declarations that you have and then apply this classification model to more parcels (but this will need some work from you on the script). So if you upload only one parcel, the L4A processor cannot be used.
For the “pa-min” parameter: it is the minimum number of parcels by crop type for the crop type to be included in the classification (after selection of parcels with min S1 and S2 pix and min S2 pix for the calibration selection). It was defined to avoid to classify very rare crop types that are so few compared to the other crop types.