v0.4.1

foldkit-durable 0.3.0 and foldkit-sync 0.4.0

• Namespaces, additively. Both packages exported bare functions while every
  other package here is namespace-first, and Effect is too. Journal.make,
  Journal.layer, Journal.metrics and Journal.define, and Sync grouping
  forApplication, define, mount, metrics, indexedDb plus transport.*,
  presence.* and lww.*. They delegate by reference, so inference at call
  sites is unchanged, and every previous export keeps its name, signature and
  behaviour. The branded ids lead with DocumentId.make('todos'), which Effect's
  schemas already provide; the lowercase decoders still work.
• Journal.define<Operation, Snapshot, Principal>(key) closes a soundness
  hole. JournalService takes its type arguments at the use site with nothing
  tying them to the layer that satisfied the tag, so reading a key back as a
  journal of another shape compiles. A definition fixes the types once and
  returns the tag and the layer constructor that agree on them.
• Effect schemas as codecs. JournalOptions.operation and .snapshot accept
  a Schema.Codec as well as the encode/decode pair, with Codec.fromSchema
  as the explicit conversion, resolved once at construction because the sync
  decoders recompile per call.
• A refusal can say why. authorize accepts { allowed: false, reason } and
  the reason reaches OperationRejectedError and its message, which matters
  because sync forwards rejections to the replica and the person whose edit
  reverted saw no reason. validate now accepts an Effect as authorize
  already did, and runEffect's retryFailed accepts a predicate over the
  record, since the recovery policy asks the application to decide per intent.