feat: cpp extension for LPC

[yoyolicoris]

C++ kernel for LPC adapted from #10. Will close #10 after merged.
Compiled with OpenMP if available.
The following is the benchmark (order=20) compared to v0.6.

[--------------------------  --------------------------]
                                |  numba_lpc  |  lpc_cpu
4 threads: ---------------------------------------------
      bs_1__n_16384__threads_4  |     341.0   |    307.1
      bs_2__n_16384__threads_4  |     343.5   |    392.9
      bs_4__n_16384__threads_4  |     374.7   |    423.4
      bs_8__n_16384__threads_4  |    2434.8   |   3099.4

Times are in microseconds (us).

The inner for-loop is not parallelised, though.
I tried some SIMD instructions with OpenMP but didn't succeed.
It should be fine since the runtime is just slightly slower than Numba.

[christhetree]

Sorry for the delay, left two comments, but the changes look good. Feel free to close my PR if you like. Maybe benchmark_forward.py could be useful from that PR, but otherwise this covers everything.