feat: add `llama.cpp` backend

[McPatate]

• add support for https://github.com/ggerganov/llama.cpp (located at examples/server)
• fix inverted condition that was causing all ranges to be invalid

[github-actions[bot]]

Repository name	Source type	Average hole completion time (s)	Pass percentage
smol-rs/async-executor	github	6.904	0.00%
jaemk/cached	github	11.474	0.00%
tkaitchuck/constrandom	github	17.101	0.00%
tiangolo/fastapi	github	27.458	29.95%
null	null	null	null%
huggingface/huggingface_hub	github	25.113	40.00%
gcanti/io-ts	github	20.485	60.00%
lancedb/lancedb	github	120.696	0.00%
mmaitre314/picklescan	github	6.164	0.00%
simple	local	6.485	0.00%
encode/starlette	github	12.268	0.00%
colinhacks/zod	github	8.601	0.00%

Note: The "hole completion time" represents the full process of:

• copying files from the setup cache directory
• replacing the code from the file with a completion from the model
• building the project
• running the tests