lookup: add print for drafting performance

[JohannesGaessler]

This PR measures the time needed to create the draft in the lookup example and prints this information at the end. As it turns out the time is negligible:

n_draft   = 4
n_predict = 1025
n_drafted = 712
t_draft   = 2.85 ms, 4.01 us per token, 249387.04 tokens per second
n_accept  = 410
accept    = 57.584%

[ggerganov]

It's better to measure the entire prompt_lookup() call:

{
    const int64_t t_start_us = ggml_time_us();

    prompt_lookup();

    t_draft_us += ggml_time_us() - t_start_us;
}