server : add slot_prompt_similarity getter/setter

[bernardladenthin]

Overview

Add public getter and setter methods for the slot_prompt_similarity field in server_context.

Changes

• Added get_slot_prompt_similarity() const method to retrieve the current threshold value.
• Added set_slot_prompt_similarity(float value) method to update the threshold value.

Additional information

This change exposes the slot-similarity configuration as part of the public server_context API, so callers can query and modify slot-selection behavior at runtime without reloading the model.

The motivation is API parity for embedders. I'm refactoring bernardladenthin/java-llama.cpp using Claude Code to drop its hand-ported server.hpp and link upstream's server_context directly. In the previous code path the JNI layer could update this threshold at runtime; with the new upstream API surface, that capability is no longer reachable, since slot_prompt_similarity lives inside the impl with no setter.

I don't have a concrete operational use case beyond restoring the previous embedder behavior, and I'm happy to drop the PR if you'd rather keep the field strictly load-time configurable. If so, it might also make sense to drop this feature on the Java-binding side. Please give an suggestion.

Requirements

• YES: I have read and agree with the contributing guidelines. I've not setup a CI because changes are minimal.
• AI usage disclosure: partial: written with Claude Code (Opus). I reviewed every line and confirmed the design against the existing on_sleeping_changed pattern.