S3 publishing dumps all assets into the root of the bucket

[sdlaver]

The uploaded files have random names, including for:

• APKs
• Off-chain JSON metadata
• image assets
• etc

This makes it hard to associate files within the bucket - you'd have to scan for which files are off-chain metadata, and then parse those for references to the other files. I'd propose splitting into subdirs based on file type, e.g.:

<bucket root>/metadata/*
<bucket root>/media/*
<bucket root>/files/*

Files would still have random names within these subdirectories.

This would provide a mechanism to easily know what is metadata, and to parse those to locate what the files in media and files correspond to.