Tiled/subset metadata distinction

[drwelby]

Placeholder for discussion on how to best represent metadata specific to a tile or other subset of a larger dataset.

Example:

A satellite pushbroom sensor collects a strip of data that is then divided into tiles. The strip could have a large amount of cloud cover, but one tile might be relatively cloud free.

In our current implementation we may process only one tile so there are no records for neighboring tiles. We represent strip-level metadata using standard EO extension tags, but then tile-specific values using a tile: prefix. This is largely because we only store metadata at the tile level.

It could be argued that the tile's metadata should only represent the tile itself, and that the strip should be represented by a STAC collection.

Since the question is likely to be repeated, I'd like to gather opinions towards a "best practice".