Truncated model files can cause llama.cpp to crash when using mmap

[slaren]

llama_model_loader does not check if the tensor data is present in the file when using mmap, and in some cases this can cause llama.cpp to crash.

To fix this, llama_model_loader should check that all the tensor data is within the bounds of the file, and otherwise stop the process and notify the user that the model file is corrupted.