feat: Inference speed(tokens/s) profiling

[yirongjie]

What's new

Introduced the module.profiling() function.
An example can be found on line 58 of example/demo_tinyllama.cpp, and the resulting output is shown below:

===========================================
  Load time: 0.461832 s
  Prefilling speed: 5.78816 tokens/s
  Decoding speed: 5.23941 tokens/s
===========================================