v8.1.0

• Changed: the web SQLite store is no longer experimental. Its last
  graduation gate, a soak, now runs in Chromium on each change to the main
  branch and on each pull request that is ready for review. It drives open,
  write, read, and close cycles, one long session of churn, and writes that
  two tabs make in turn, and it fails when latency, memory, or the file size
  grows over the run. The guides now name webSqliteStore first for
  browsers where the origin private file system works, and indexedDbStore
  where it is not available. The error of DefaultSignalProtocolClient.create()
  without storage names both.
• Fixed: an open of the web SQLite store under a full origin quota rejects
  with StorageQuotaExceededError. The open rejected with
  OPFS_FILE_BUSY. After a failed first open, the store could also fail every
  later open with SQLITE_CANTOPEN. Now the open rejects with
  StorageQuotaExceededError (STORAGE_QUOTA_EXCEEDED), a new store writes
  no key and no database file, and the open succeeds after space frees.
• Changed: hosted prekey reads send the destination mailbox generation.
  The hosted Relay client sends the mailboxGeneration field on each prekey
  bundle fetch and each prekey status read, which includes an identity key
  read. The Relay keeps the prekeys of a device for each mailbox generation,
  so a read after a recovery gets the prekeys of the recovered device. The own
  device sends the generation of its session. For another device, the client
  reads the device directory first. When the Relay does not find the prekeys
  at a cached generation, the client reads the directory one more time and
  tries again. A directory read now also removes the cached generation of each
  device that the directory no longer lists.
• Changed: group zero-knowledge proofs and credential presentations are
  about twice as fast. The constant generators of the group credential and
  encryption systems keep a table of their multiples and their encoding. A
  proof, a presentation, and a presentation verification calculate each sum
  of products with one constant-time multiscalar multiplication, which uses
  these tables. The Lizard decode uses an addition chain for its field
  exponentiation. The proofs, the credentials, and their bytes do not change.