Clarification on the Internal Structure of a COG (Cloud Optimized GeoTIFF) and the _ifds Function

Nobody has claimed this yet.

I am trying to better understand how [COGTiff class](https://github.com/mapbox/COGDumper/blob/master/cogdumper/cog_tiles.py) works like whats the internal structure of a Cloud Optimized GeoTIFF (COG). Specifically, I'm looking at the _ifds function and have a few questions about the way it works.

From my understanding, the function recursively reads the Image File Directories (IFDs) in the COG, and I assume that the structure of the COG is closely tied to the "virtual VRT" (Virtual Raster Tile) format. It seems like the function uses this structure to determine the start and end memory locations of tiles based on their x, y, and z coordinates.

Could you provide more insight into how the internal structure is defined in a COG? Are there specific references or documentation that explain how the "virtual VRT" structure is used to handle tiles, especially in terms of how the offsets and byte counts are calculated and stored?

Start by reading the COGTiff class and its _ifds function in cogdumper/cog_tiles.py. Compare the function's handling of IFDs, offsets, byte counts, and tile coordinates with the COG and virtual VRT references identified during research; done means the repository has a clear explanation or reference for these relationships.

Written by the indexing model from the issue text.