test(hosting): close long-tail package coverage gaps

Goal

Close coverage gaps in maintained packages which are individually smaller than the primary workstreams but collectively block the repository from exceeding 90%.

Coverage report: https://github.com/dotnet/orleans/runs/98679059570

Initial scope

Representative physical-source gaps include:

• Dashboard: roughly 540 uncovered lines and near 50% coverage.
• Connections security: roughly 230 uncovered lines and near 46% coverage.
• Hosting, diagnostics, observability, dependency injection, reminders, memory persistence, clustering providers, and serializer adapters have smaller but material gaps.
• Dashboard.ServiceCollectionExtensions.cs is near 18% coverage.
• TLS/duplex-pipe failure paths, dashboard service registration/options, metrics/diagnostic events, and hosting configuration branches are under-covered.

Work

• Produce a package-by-package table of covered/valid lines and branches after the larger area issues land.
• Prioritize packages below 80%, then close remaining gaps needed to exceed the repository goal.
• Add dashboard registration, options, authorization, routing, and failure-path tests.
• Add connection-security handshake, stream contract, cancellation, certificate, and error-path tests.
• Cover diagnostics/metrics emission and disabled/configured behavior.
• Cover hosting and dependency-injection validation, named options, duplicate registration, and configuration failures.
• Cover small reminder, persistence, clustering, and serializer-adapter branches using existing fixtures.
• Split implementation PRs by package/owner while tracking them in this issue.

Acceptance criteria

• Every maintained package is at least 90% line coverage, or has an evidence-backed exception tracked in the dead-code/reachability audit.
• No package below 80% remains unassigned.
• Tests assert runtime and configuration outcomes rather than only resolving services.
• Generated files, reference API surfaces, and platform-specific code are treated consistently by the canonical coverage model.