Saving and restoring a workspace snapshot hashed, encrypted and decrypted
every byte on the server's only thread, and so did the Content-MD5 of each
uploaded part. A 1 GiB save kept that thread busy for about 2.9 s.
These passes now run on at most two worker threads
(server/src/storage/file-crypto-pool.ts). A job names files and byte
ranges, so a waiting job holds no data, and a cancelled job stops at its
next chunk. Every check runs as before, in the same order, with the same
failures. The stored format is unchanged.
A restore reads the decrypted archive once instead of twice: it is digested
while it is copied beside its destination, and linked only when the digest
matches.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>