fmriprep does only produce “minimally preprocessed” outputs.
It is up to the user to load a sensible subset of confounds generated by fmriprep and apply them.
fmriprep does generate a basis of discrete cosines that can be used for high pass filtering, but I don’t think it has the basis for the low pass filtering.
However this is easy to implement with nilearn's NiftiMasker. You just to be careful how you include your confounds in the masker.
Just to add on to @pbellec’s helpful response: filtering, along with other denoising steps like nuisance regression, are typically the last preprocessing steps you do. That means a standard pipeline (e.g. for functional connectivity) may be something like fMRIPrep → quality control (e.g. exclude if too much motion or poor spatial normalization) → spatial smoothing (several software packages like SPM and Nilearn implement this) → denosing (filtering, nuisance regression, volume censoring etc.) → 1st and 2nd level analyses.
fMRIPrep will get you through the first two steps, since you can use the HTML outputs of fMRIPrep and the confounds file to check for poor spatial normalization and motion outliers, respectively.