Fix small data discrepancy between Xee and pure computePixels()

[alxmrs]

Description TBD.

[alxmrs]

Note for the reader: if you are just using Xee, this is not so much of an issue. But, if you need to replace a raw computePixels() call with Xee, this bug will be a concern.

[alxmrs]

This is the internal test that fails:

class GetPixelsTest(unittest.TestCase):

  def setUp(self):
    super().setUp()
    init_ee_for_tests()

    self.lnglat = ee.ImageCollection.fromImages([ee.Image.pixelLonLat()])
    self.conus = ee.ImageCollection('GRIDMET/DROUGHT').filterDate(
        '2020-03-30', '2020-04-01'
    )
    self.bbox = ee.Geometry.BBox(-121.55, 39.01, -120.57, 39.38)

  def test_get_pixels__with_or_without_xee__is_the_same(self):
    for name, ic in [
        # ('latlng', self.lnglat),  # TODO(alxr, mahrsee): Support this case.
        ('conus', self.conus),
    ]:
      img = ic.first()
      with self.subTest(f'{name} test'):
        clipped = img.clip(self.bbox)

        y = ee_util.get_pixels(clipped, using_xee=False)
        y_hat = ee_util.get_pixels(clipped, using_xee=True)

        self.assertEqual(y.shape, y_hat.shape, f'{y.shape} != {y_hat.shape}')
        self.assertEqual(y.dtype, y_hat.dtype, f'{y.dtype} != {y_hat.dtype}')
        # np.array_equal better compares structured arrays than np.allclose
        self.assertTrue(np.array_equal(y, y_hat), f'{y}\n!={y_hat}')

It fails with the following error:

Traceback (most recent call last):
  File ".../ee_util_manual_test.py", line 74, in test_get_pixels__with_or_without_xee__is_the_same
    self.assertEqual(y.shape, y_hat.shape, f'{y.shape} != {y_hat.shape}')
AssertionError: Tuples differ: (10, 25) != (9, 24)

First differing element 0:
10
9

- (10, 25)
+ (9, 24) : (10, 25) != (9, 24)

[alxmrs]

After talking to @saicheems and other Earth Engine folks, it seems like the root of the issue has to do with rounding the pixel boundaries on the backend that is not mirrored in this Python client.

[mahrsee1997]

Resolved in #105.