Better VFS integrity and syncing

[james-pre]

#286 lead to me going down a rabbit hole and realizing there is a whole collection of issues with the VFS with regards to data integrity and sync behavior:

• Multiple open Handles for the same file don't share metadata #287, meaning the inode can get out of sync
• Caching the inodes means that VFS stat needs to go through the cache first
• Handle.write() does not update inode metadata atomically #286 (writes to the backend immediately but waits to sync metadata until being closed)
  • Proper atomic wrties means caching written data before syncing it. Caching in chunks (i.e. a page cache) is necessary to avoid reading the entire file into memory for an arbitrarily small write
  • unsynced writes also needs to be shared across open Handles