[Bug] Instantaneous rate estimation fails for inconsistent kernel selection

[JuliaSprenger]

When providing a kernel on a much smaller time scale than sampling rate requested the instantaneous rate estimation will fail on numpy level with
ValueError: zero-size array to reduction operation minimum which has no identity

Here's a minimal failing example

from elephant.statistics import instantaneous_rate
from elephant import kernels
spiketrain = neo.spiketrain_generation.homogeneous_poisson_process(10*Hz, t_start=array(0.) * ms, t_stop=array(10000.) * ms)
kernel = kernels.AlphaKernel(sigma = 5*ms, invert = True)
rate = instantaneous_rate(spiketrain, sampling_period = 200*ms, kernel=kernel)

I would expect this to be caught by the instantaneous rate and providing a more telling error message instead of this failing on a numpy level.

The full traceback is

ValueError                                Traceback (most recent call last)
<ipython-input-167-5ac422ce9801> in <module>
      1 from elephant import kernels
      2 kernel = kernels.AlphaKernel(sigma = 5*ms, invert = True)
----> 3 rate = instantaneous_rate(spiketrain1, sampling_period = 200*ms, kernel=kernel)

~/anaconda3/envs/t/lib/python3.7/site-packages/elephant/statistics.py in instantaneous_rate(spiketrain, sampling_period, kernel, cutoff, t_start, t_stop, trim)
    743 
    744     if not trim:
--> 745         r = r[kernel.median_index(t_arr):-(kernel(t_arr).size -
    746                                            kernel.median_index(t_arr))]
    747     elif trim:

~/anaconda3/envs/t/lib/python3.7/site-packages/elephant/kernels.py in median_index(self, t)
    215         """
    216         return np.nonzero(self(t).cumsum() *
--> 217                           (t[len(t) - 1] - t[0]) / (len(t) - 1) >= 0.5)[0].min()
    218 
    219     def is_symmetric(self):

~/anaconda3/envs/t/lib/python3.7/site-packages/numpy/core/_methods.py in _amin(a, axis, out, keepdims, initial)
     30 def _amin(a, axis=None, out=None, keepdims=False,
     31           initial=_NoValue):
---> 32     return umr_minimum(a, axis, None, out, keepdims, initial)
     33 
     34 def _sum(a, axis=None, dtype=None, out=None, keepdims=False,

ValueError: zero-size array to reduction operation minimum which has no identity

[dizcza]

1. Are you sure we should raise an error in this case?
2. Is t in median_index() supposed to be sorted?
3. A median index is something undefined unless t is sorted. Even if it's sorted, it's still undefined for arrays with an even number of values. Can we rewrite the logic such that a median, not its index, is returned? Then we don't need to tackle this issue.

[JuliaSprenger]

Regarding 1. Of course raising a warning instead and providing Nan as return value is also an option. I would argue again introducing an automacigal adjustment of the input parameters as this makes the function less transparent.
Regarding 2 & 3 I don't know / have a strong opinion.