here's my take on defining a "geobox" at a longlat point, where I only need a longitude, latitude, distance, and n-pixels
https://gist.github.com/mdsumner/2af156a8250dd825221c634f1b6fdeba
I use a local projection to ease the geographic extent definition, then transform that to (auto) UTM, then get the longlat bbox (for STAC query), and the UTM GeoBox (for odc to render to)