Fix bug in GaussianFilter

[DradeAW]

Following up on the discussion we had in #2397

Indeed there is a bug in the Gaussian filter implementation, which matters if freq_min and freq_max are close to one another, or even if the min is above the max

Here is a quick code to check for the bug:

import numpy as np
import plotly.graph_objects as go
import spikeinterface.core as si
import spikeinterface.preprocessing as spre


sf = 30_000.0


# An impulse contains all frequencies equally.
trace = np.zeros([int(sf), 1], dtype=np.float64)
trace[15000] = 1.0

recording = si.NumpyRecording(trace, sf)
recording_f = spre.gaussian_filter(recording, freq_min=600, freq_max=300)

trace_f = recording_f.get_traces()


fig = go.Figure()

fig.add_trace(go.Scatter(
	x=1e3 * np.arange(sf) / sf,
	y=trace[:, 0],
	mode="lines",
	name="Original trace"
))
fig.add_trace(go.Scatter(
	x=1e3 * np.arange(sf) / sf,
	y=trace_f[:, 0],
	mode="lines",
	name="Filtered trace"
))

fig.update_xaxes(title_text="Time (ms)")

fig.show()

Here is what the plot looks like before the patch:

And what is looks like after this patch:

We can see the impulse is no longer going in the other direction, and there is less signal overall (which is expected)
Thanks @TomBugnon for pointing out the issue!