What does NoInfer<T> do and when should you block inference?

Question presented to candidate: "TypeScript 5.4 added a utility type called NoInfer. What problem does it solve, how does it change type inference, and can you show me a function signature where you would actually reach for it?"

What a strong answer should cover:

• Every occurrence of a type parameter in the parameter list is an inference site; candidates from all sites are combined, so a "secondary" argument (a default, a fallback, a transition target) can silently widen T
• NoInfer<T> (TypeScript 5.4) marks one position as "check only": T is inferred from the other sites first, then the argument in the NoInfer position is checked against it
• The classic example: light<C extends string>(colors: C[], defaultColor?: NoInfer<C>) rejects "blue" instead of adding it to C
• Before 5.4: an extra type parameter D extends C, or a deferred conditional trick; C & {} does not block inference
• Edge cases: NoInfer<X> on a concrete type is just X; if NoInfer is the only site, T falls back to unknown or its default; it is erased at run time
• When not to use it: when every argument should legitimately contribute, or when the plain signature already reports the mistake

Code / implementation expected: Yes — a generic function where one argument must conform to a type inferred from another, with and without NoInfer.