llama.cpp Release b11944 Adds Chat Template Reasoning Effort Levels

Release b11944 for llama.cpp introduces server-side exposure for reasoning effort levels within chat templates. The update includes a capability probe that tests various candidate levels—ranging from minimal to maximum—and reports the supported levels via the server's properties endpoint. This feature facilitates better integration and control over advanced reasoning models that require dynamic thinking depth configuration. Users running local large language models can now fine-tune the reasoning behavior directly through the server API. The capability probe evaluates candidate effort levels (minimal, low, medium, high, xhigh, max) by rendering user turns and detecting fallbacks. The resulting supported levels are serialized into JSON and exposed through the chat_template_caps field in the /props endpoint.