Update to latest llama.cpp

[philpax]

There have been quite a few changes since our last major sync: https://github.com/ggerganov/llama.cpp/compare/904d2a8d6acd667c9633138d45a361d40fbf76d0..HEAD

(There may be others we haven't accounted for in the inferencing code).

Need to do a more precise breakdown, but

• llama-cli: Could not load model: InvalidMagic { path: ... } #59 and Support latest (breaking) changes to GGML file format #61

[iacore]

I met the InvalidMagic error again, after #61. Is there another breaking change in model format?

[philpax]

Yep, #93.