Create the actor system from plain values (cluster name as a string).

Register an aggregate and return its typed handle. Calling this IS the registration (it initializes the sharding region).

An aggregate id from a string (e.g. a document/user key). Any non-blank id works: the shard names entity actors Uri.EscapeDataString(entityId), so characters Akka actor names would reject directly (spaces, %) are escaped before they reach an actor path.

Register an aggregate whose `decide` uses `dispatch` (the RunAsync effect), supplying the runner that turns an effect DESCRIPTION into a command sent back to the aggregate. `decide` stays pure; the oracle or side effect lives only here. The runner MUST be total (wrap it with `total` so an oracle error becomes a command, never an escaping exception). See `EventAction.RunAsync` for the ephemeral contract.

A correlation id from a string you already have (a request id, a trace id). Rejects '~': the separator FCQRS builds saga entity names and pub-sub topics from, so a CID containing one is parsed back wrong and leaves the saga permanently deaf. `newCid` needs no such check — a UUID has none.

Build a SQLite/etc. Connection from a raw connection string (ShortString hidden).

 Register stable journal names for payload types (see JournalTypes):
     Fcqrs.journalTypes [ journalType "doc.event"; ... ]
 Call before the actor system writes anything.

A fresh correlation id (UUID v7).

Register the read-model projection and return the subscription stream.

Register a saga and return its handle. The originator-event type is inferred from `def.StartOn`, so there are no type arguments to supply: `Fcqrs.saga api def`.

Read-your-writes in one call: subscribe on the CID BEFORE sending, send, then await the projection ONLY if the delivered ack was journaled. A deferred (rejection-style) ack never reaches the journal, so a naive await would hang until timeout. The subscribe-before-send ordering is what makes the wait race-free; owning it here means callers cannot get it backwards. Awaits exactly ONE projected event: a batch persist (PersistAllEvents) caller should Subscribe with an explicit take instead. Envelopes without the delivery stamp (pre-stamp FCQRS) are treated as journaled. The projection wait is bounded by `akka.fcqrs.command-timeout` (default 30s, bare number = seconds): a projection that suppresses or filters out the matching notification raises TimeoutException instead of hanging the caller forever.

Wire every registered saga into one saga-starter (or the empty starter if none). Call after the aggregates + sagas are registered.