generator: llama/gguf

[leondz]

ggml is an initiative to build small models through quantisation of parameters to smaller datatypes (ggml.ai)

Garak supports an old version of their interface, ggml files + llama.cpp

Things have move forward with ggml. The llama/ggml connector in garak seems to be problematic/broken now, and also, a new gguf file format has emerged.

We should

1. support gguf
2. fix or remove the existing ggml file interface
3. consider having some kind of test for this, depending on the degree of orchestration required

[leondz]

see #540 , #474