Include gpt2_update in timings of C code

[nietras]

@karpathy in my timings for cpu the update step is a non-neglible part of the training time, should this not be included? E.g. for my Llm.cs port I get timings like (on AMD 5950X):

 0: loss 5.269892 exp. 5.270007 OK   ( 3700 ms = Forward  1043 ms ZeroGrad   0 ms Backward 2091 ms Update  566 ms)
 1: loss 4.059389 exp. 4.059707 OK   ( 3494 ms = Forward   919 ms ZeroGrad  28 ms Backward 2039 ms Update  508 ms)

note how update step here is ~500 ms out of ~3500 ms of total. (This is after a few optimizations). Update step should be fairly easy to optimize via openmp I'd guess for C version.