Shift in dem.value_at_coord or dem.show

[LucBeraud]

Describe the bug
There seems to be a diagonal ~ half-pixel shift between the DEM.show() displays and the DEM.value_at_coord() output.

To Reproduce
import xdem
img = xdem.DEM(gu.examples.get_path('everest_landsat_b4'))
target_x, target_y = 486811, 3097050 # example of coordinate
fig,ax=plt.subplots(1,1)
img.show(vmin=50, vmax=200, ax=ax)
ax.set_xlim(target_x-60,target_x+60); ax.set_ylim(target_y-60,target_y+60) # zoom on the area
ax.scatter(target_x, target_y,c='r') # plot the desired coordinate on the image plot
ax.set_title('at ({}, {}), dem.value_at_coords={}'.format(target_x, target_y, np.round(img.value_at_coords(target_x, target_y), 2))) # prints as figure title the value returned by value_at_coords()
ax.plot([486805, 486805, 486805+30, 486805+30, 486805],[3097056,3097056-30, 3097056-30, 3097056, 3097056]) # if coord falls in this blue polygon
ax.plot([486820, 486820, 486820+30, 486820+30, 486820],[3097040,3097040-30, 3097040-30, 3097040, 3097040]) # it returns the value of this orange one

Expected behavior
The value_at_coord() value does not return same value as the plot most of the time, I expect that the value_at_coords() (169 in the example, value of the orange box pixel) matches the show() pixels (131 in the example, under the red point).

System (please complete the following information):

• OS: Ubuntu 22
• Environment : xdem (0.0.7, 823fff2 commit), geoutils (0.0.10)
  

[rhugonnet]

Thanks for the reproducible example, helped a lot!

Found the guilty line: 3267726

Turns out our test_value_at_coords function had the same value for neighbouring pixels, and so it went unnoticed. This is fixed as well: 147363e

Will be closed when #351 is merged.