Model resolution now distinguishes explicit vs. persisted spelling

What

• The startup_resolve_model telemetry event, which previously always logged family_spelling_unserved when a model name resolves to nothing despite a raw request existing, now logs persisted_spelling_unserved instead in that case.
• Model-resolution helpers now pass an explicit explicit or persisted tag through a renamed function (XUt, previously sFt), and a companion helper checks whether a model is valid or comes from a persisted source.

Why Distinguishing whether an unresolved or unusual model name was typed explicitly versus loaded from persisted settings makes it possible to tell these two situations apart in telemetry and resolution logic, rather than lumping them together as one generic case.