adds `align_bounds_with_dataset` to adjust bounds in reader.part() method

[vincentsarago]

Thanks to @abarciauskas-bgse for deep-diving into this issue.

When doing zonal statistics, users might not want to get the stats from resampled data but clip the data to the closest rounding window (full pixel) and then use the coverage option within the statistics method to adjust the values.

With this PR we're introduction an option in rio_tiler.reader.part to adjust the bounds to the dataset transform. This option might be used from feature() orpart()` reader's method such as

with Reader("cog.tif") as src:
    data = src.feature(
        shape,
        shape_crs=WGS84_CRS,
        align_bounds_with_dataset=True,
    )

    coverage_array = data.get_coverage_array(
        shape,
        shape_crs=WGS84_CRS,
    )

    stats = data.statistics(coverage=coverage_array)