Compute TFR with fmin=0Hz

Currently, tfr_morlet and tfr_multitaper throw an error if the minimum frequency is 0Hz:

/usr/local/lib/python3.7/site-packages/mne/time_frequency/tfr.py in morlet(sfreq, freqs, n_cycles, sigma, zero_mean)
     83         # this scaling factor is proportional to (Tallon-Baudry 98):
     84         # (sigma_t*sqrt(pi))^(-1/2);
---> 85         t = np.arange(0., 5. * sigma_t, 1.0 / sfreq)
     86         t = np.r_[-t[::-1], t[1:]]
     87         oscillation = np.exp(2.0 * 1j * np.pi * f * t)

ValueError: arange: cannot compute length

and

/usr/local/lib/python3.7/site-packages/mne/time_frequency/tfr.py in _make_dpss(sfreq, freqs, n_cycles, time_bandwidth, zero_mean)
    140 
    141             t_win = this_n_cycles / float(f)
--> 142             t = np.arange(0., t_win, 1.0 / sfreq)
    143             # Making sure wavelets are centered before tapering
    144             oscillation = np.exp(2.0 * 1j * np.pi * f * (t - t_win / 2.))

ValueError: arange: cannot compute length

FWIW, tfr_stockwell works fine with 0Hz.

Maybe I'm totally misunderstanding something, but I think it would be nice if TFRs started at DC, independent of the TFR method.

MWE to reproduce:

import numpy as np
import mne

sfreq = 128
n_epochs, n_ch, n_time = 144, 2, 1921
data = np.random.randn(n_epochs, n_ch, n_time)
info = mne.create_info(n_ch, sfreq, 'eeg')
epochs = mne.EpochsArray(data, info)

freqs = np.arange(0, 4)
mne.time_frequency.tfr_morlet(epochs, freqs, n_cycles=freqs / 2)  # throws error
mne.time_frequency.tfr_multitaper(epochs, freqs, n_cycles=freqs / 2)  # throws error
mne.time_frequency.tfr_stockwell(epochs, fmin=0, fmax=3)  # works

Start by reproducing the MWE with tfr_morlet and tfr_multitaper, then inspect the wavelet code in tfr.py at the failing arange calls for zero frequency. Compare their behavior with tfr_stockwell, which already accepts fmin=0. Done means both methods compute a TFR starting at 0 Hz without the reported error.