a_star_search maps out-of-bounds points to interior pixels and truncates instead of rounding in _get_pixel_id

[brendancol]

Describe the bug

_get_pixel_id() in xrspatial/pathfinding.py converts a (y, x) point to a pixel index with

py = int(abs(point[0] - y_coords[0]) / cellsize_y)
px = int(abs(point[1] - x_coords[0]) / cellsize_x)

Two problems come out of this:

1. abs() folds out-of-bounds points back inside the raster. A point past the edge on the coords[0] side gets a positive offset and maps to a mirrored interior pixel. a_star_search then runs from the wrong start pixel instead of raising the "start location outside the surface graph" error it raises for points past the opposite edge. multi_stop_search's waypoint bounds check has the same blind spot.

2. int() truncates instead of rounding to the nearest cell center. Any point in the upper half of a cell is assigned to the neighboring lower-index cell (a half-cell bias), and a point that should land exactly on a cell center can flip to the previous cell through float noise: on a grid with 0.1 resolution, x=0.3 maps to pixel 2 because the pixel-3 center is 0.30000000000000004.

Reproduction (run on this host):

import numpy as np
import xarray as xr
from xrspatial import a_star_search
from xrspatial.pathfinding import _get_pixel_id

h = w = 5
agg = xr.DataArray(np.ones((h, w)), dims=['y', 'x'], attrs={'res': (1.0, 1.0)})
agg['y'] = np.linspace(h - 1, 0, h)   # [4, 3, 2, 1, 0]
agg['x'] = np.linspace(0, w - 1, w)

# 1) start y=5.0 is OUTSIDE the raster (top edge is y=4.0)
path = a_star_search(agg, (5.0, 0.0), (0.0, 4.0))
print(np.argwhere(path.values == 0))   # -> [[1 0]]  path starts at row 1, no error

# 2) truncation
agg2 = xr.DataArray(np.ones((2, 10)), dims=['y', 'x'])
agg2['y'] = [0.1, 0.0]
agg2['x'] = np.arange(10) * 0.1
print(_get_pixel_id((0.0, 0.3), agg2, 'x', 'y'))   # -> (0, 2), pixel-3 center is 0.30000000000000004
print(_get_pixel_id((0.0, 2.6), agg, 'x', 'y'))    # -> (?, 2) on a unit grid, nearest center is 3

Observed output:

[[1 0]]
(0, 2)
(2, 2)

Expected behavior

• A point outside the raster bounds raises ValueError ("start location outside the surface graph") regardless of which side it falls on.
• A point maps to the cell whose center is nearest, so x=0.3 on a 0.1-resolution grid resolves to pixel 3 and x=2.6 on a unit grid resolves to pixel 3.

The fix is to compute a signed offset using the coordinate direction and round to the nearest index; negative indices then fail the existing _is_inside check and raise as intended.

Additional context

Found by the accuracy sweep (Cat 1/Cat 3). The folding half is rated HIGH (silent wrong path from a mis-mapped start/goal), the truncation half MEDIUM. Both a_star_search and multi_stop_search route through _get_pixel_id. Existing tests only use points that sit exactly on cell centers, so they pass under either convention.