Pixel discarding for f32 should use float next after for precision issues

If the nodata value is defined as a float64, then we need to consider precision issues because the GPU only supports float32.

We should be able to vendor https://github.com/scijs/nextafter/blob/master/nextafter.js. Then for filtering we may need to filter inside a range of `nextAfter(val, -Infinity), nextAfter(val, Infinity)`

Start with the referenced nextafter.js implementation and trace the pixel filtering path that compares float32 GPU values with float64 nodata values. The work is done when the next-after bounds are used for filtering around the nodata value and precision-related pixel discarding is covered.

Written by the indexing model from the issue text.