r.horizon: variable needs to be reset for point mode

[petrasovaa]

trafficstars

Another bug coming from using global variables in r.horizon. Reported by a user, can be reproduced with https://www.datigeo-piem-download.it/static/regp01/DTM5_ICE/RIPRESA_AEREA_ICE_2009_2011_DTM-SDO_CTR_FOGLI50-172-EPSG32632-TIF.zip:

r.horizon elevation=DTM5_172 direction=0 step=1 coordinates=350089,4979561 -d file=horizon.csv

The bug is that for part of the angles it results in zeros.

I am not sure how to trigger the problem besides this specific DEM, so I didn't add any test.

This bug is not present in #3346, there it was (accidentally) fixed.