fix(llama.cpp): include model name in completion request body (#11803)

The llama.cpp provider was not sending the `model` field in the
request body, causing 400 errors ("model name is missing from the
request") when using llama-server's model router feature.

Fixes #11244
Fixes #10846