Prompt sources

Prompts showed how a deployment substitutes what a key produces with Prompts.replace(key, source) or puts one source behind every key with Prompts.setGlobalBackend(source). Either way it needs a PromptSource, and the usual needs are the same everywhere: text from a database, text from a server, an experiment between two wordings. This package ships those, so a deployment plugs one in instead of writing a source class of its own.

• DbTextSource looks the content up by key through a Lookup function you hand it. The function is yours, so your session, transaction or DAO stays where it lives and Nucleo needs to know nothing about it.
• RemoteTextSource fetches the content by key through a Fetcher you hand it, and the HTTP mechanics stay yours the same way.
• AbTestSource answers from one of two inner sources, choosing by probability: probA between 0 and 1, checked at construction, decides how often the first branch answers. Wrap the current wording and the candidate wording, and the traffic splits without a release.
• StaticTextSource is a fixed string captured at construction. You rarely build one yourself: it is what the scanner mints for every @StaticPrompt string constant.

StaticTextSource carries the StaticPromptSource marker, so a prompt produced from it is kept after its first check. The other three answer per call, since their content can change under the key; the guardrail verdicts are remembered by content, so a check repeats only when the content actually changed.