[examples][deepseek] Correct and improve timing for model execution

[ArCyanic]

The origin timing is wrong, this commit fixed that and added extra significant information for profiling.
Now it looks like this:

[Total time] 332.403s
[Prefill time] 36.8335s
[KV cache copy time] 7.6678e-05s
[Decode time] 295.569s
[Prefilling] 0.271492 tokens/s
[Decoding] 2.85213 tokens/s