MFM_DPFA_clean

clean & optimize the code

update notes:

-06/02/2022: clean the code for labeled data. 😊

1. use C code for PGdraw;
2. replace the user-defined NUTS with MATLAB built-in HMC;
3. pretty fast with (1) & (2);
4. BDMCMC works well, infer (p=2) correctly (just check first 1000 iterations).

// TODO:

1. parallel computing; (done)
2. clean the code for clustering;
3. double check the split-merge: should change after using PG-FFBS-MH;
4. If can easily do, write FFBS in C.