Raise error when peaks overlapping in merge_peaks

[dachengx]

This PR just adds a safeguard.

If we do need to handle the case where peaks overlap, we need to change two lines:

1. strax/strax/processing/peak_merging.py

   Line 44 in dd62b13

   new_p["length"] = (strax.endtime(last_peak) - new_p["time"]) // common_dt

   , we can not assume last_peak has the largest endtime

2. strax/strax/processing/peak_merging.py

   Line 80 in dd62b13

   buffer[i0 : i0 + n_after] = np.repeat(p["data"][: p["length"]], upsample) / upsample

   , should change = to +=

[coveralls]

coverage: 89.323% (-0.8%) from 90.074%
when pulling 97e1de9 on merge_peaks_overlap
into dd62b13 on master.