Stratara.ServiceDefaults 4.0.3

• net10.0

  • Destructurama.Attributed (>= 5.3.0)
  • Microsoft.Extensions.Http.Resilience (>= 10.9.0)
  • Microsoft.Extensions.ServiceDiscovery (>= 10.9.0)
  • OpenTelemetry.Exporter.OpenTelemetryProtocol (>= 1.18.0)
  • OpenTelemetry.Extensions.Hosting (>= 1.18.0)
  • OpenTelemetry.Instrumentation.EntityFrameworkCore (>= 1.18.0-beta.1)
  • OpenTelemetry.Instrumentation.Http (>= 1.18.0)
  • OpenTelemetry.Instrumentation.Runtime (>= 1.18.0)
  • RabbitMQ.Client.OpenTelemetry (>= 1.0.0-rc.2)
  • Serilog.Extensions.Hosting (>= 10.0.0)
  • Serilog.Formatting.Compact (>= 3.0.0)
  • Serilog.Settings.Configuration (>= 10.0.1)
  • Serilog.Sinks.Async (>= 2.1.0)
  • Serilog.Sinks.Console (>= 6.1.1)
  • Serilog.Sinks.File (>= 7.0.0)
  • Serilog.Sinks.OpenTelemetry (>= 4.2.0)
  • Stratara.Shared (>= 4.0.3)

Three fixes to the same complaint: a host composed exactly as the documentation describes did not
start. Each one was a registration the composites assumed somebody else had made — a tracer, a unit
of work, a Redis connection — and each is now made by the registration that needs it. Nothing that
already worked changes: every fix defers to a registration the host makes itself, in either order,
so the lines a working host added to get past these gaps can simply be deleted. Additive on every
published surface.

### Fixed

- **A host without Redis starts and dispatches.** Every composite that carries a dispatcher
 registered the projection-replay state, and its only implementation took a Redis connection that
 no composite registers, so a host composed as documented failed at its first dispatch unless it
 also ran Redis and called `AddCaching()`. `AddProjectionReplayState()` now chooses at first
 resolution: with a registered `IConnectionMultiplexer` the Redis-backed state as before, without
 one an in-process state with the same lease semantics — and a warning, once at start-up
 (`104_012`), that replay coordination is confined to that process, so a replay requested there
 suppresses publication there only. A deployment whose replay must reach several hosts registers
 the shared connection, in either order. Redis-backed hosts observe no change.
- **Registering a store context now registers its unit of work.**
 `AddNpgsqlWriteDbContextFactory<T>()` registered the context factory, the context and the default
 connection resolver but not the `IWriteUnitOfWork` that the event source, the outbox dispatcher and
 the command worker take from the container — and nothing else in the published packages did, so a
 host composed exactly as documented failed at its first command with a dependency-injection error.
 The write registration now try-adds a scoped `IWriteUnitOfWork` over its context, and
 `AddNpgsqlReadDbContextFactory<T>()` try-adds `IProjectionsUnitOfWork` and `IReadUnitOfWork`
 likewise. A unit of work the host registers itself, before or after, is still the one used; a
 hand-written registration can simply be deleted.
- **`AddMediator()` no longer requires the host to register an OpenTelemetry `Tracer`.** The
 mediator traces every dispatch and obtained its tracer from the host, but nothing registered one,
 so a host that called `AddMediator()` and nothing else failed at the first resolve of `IMediator`.
 `AddMediator()` now registers a fallback that emits the dispatch spans from the framework's
 `Stratara.Application` activity source — a host that subscribes to framework telemetry sees them,
 a host that subscribes to nothing pays for nothing. A `Tracer` the host registers, before or after
 `AddMediator()`, is still the one used, so no existing host changes behaviour; the registration
 line can simply be deleted. The samples, the README and the package README no longer carry it.