Non-integer oversampling factors cause index error in TLS pipeline

[e-gillis]

When the Oversampling factor is set to a float with more than one decimal place, or a float less than 1, TLS fails because of an index error in the running_median function because a non-integer kernel is passed into running_median.

This happens when running_median is called through the spectra function, and can be reproduced by running:

from transitleastsquares import transitleastsquares
import numpy as np

# Random noise timeseries
t = np.linspace(0, 30, num=300)
data = np.random.normal(size=t.shape)*0.1 + 1

model = transitleastsquares(t, data)
# Any oversampling factor with more than two decimal places will cause an error
result = model.power(oversampling_factor=3.14)

Adding something like kernel = int(kernel) to the spectra function to ensure that the median filter kernel is an integer should fix this issue.

After submitting, I'm seeing that this was already discussed in the follow-ups to https://github.com/hippke/tls/issues/88, but it seems the bug is still outstanding.

Install Details:
MacOS, Python 3.10.7, TLS Version 1.0.31