Add support for response-specific meta-d'

[kevingoneill]

Currently, the hmetad package estimates a single M-ratio for both responses. Although it is less commonly used, Maniscalco & Lau (2014) showed that a more flexible approach is to use different M-ratios for the two responses. This could in principle be supported in the hmetad package, although it would require:

1. Modifying metad and stancode_metad to include response-specific M-ratio parameters
2. Modifying fit_metad to include argument for response-specific M-ratio parameters
3. Modifying posterior_epred_metad, log_lik_metad, and posterior_predict_metad to detect & use response-specific M-ratio parameters
4. Modifying _draws and _rvars functions to use response-specific M-ratio parameters