so, i remembered what the problem with chunking is. for each chunk, it stores a uri containing the txid to recover the chunk. that turns into a _lot_ of uris, which seriuosly bloats the git-annex branch and slows everything.
my intuition was that it was doing that too much to sustain. i needed to turn off the chunking. so, that seems important. the gitlakepy backend does subchunking that doesn't need local storage to index, and can be verified in depth with an fsck on the remote.
