llama internally supports load cancellation, but when using common* API it's not available from outside:

struct common_init_result common_init_from_params(common_params & params) {
    common_init_result iparams;
    auto mparams = common_model_params_to_llama(params);

    llama_model * model = llama_model_load_from_file(params.model.path.c_str(), mparams); // <<< ----
    if (model == NULL) {
        LOG_ERR("%s: failed to load model '%s'\n", __func__, params.model.path.c_str());
        return iparams;
    }
[...]

to allow load cancellation it needs a progress_callback, but it requires passing common_params and internally llama_model_params builds the object always with progress_callback as null.

this PR simply provides an opportunity to set it from outside. however I called it load_progress_callback because the callback is used during model loading. user will still need to call llama_set_abort_callback for compute cancellation, but it can be done after calling common_init_from_params().