Terminate statistical analyses for unsupported mixed data types

[neurolabusc]

The most recent release of dcm2niix changed its default behavior to preserve original data type (e.g. the default is now -l o, where old default behavior was -l n). This ensures that multiple runs acquired with UINT16 datatype are always saved as UINT16 (which AFNI does not internally support, so they get promoted to FLOAT32 at the cost of disk space). This avoids the rare situation where a first level analysis might have two series of data where one run was saved as the supported INT16 and the other as FLOAT32, with AFNI generating bogus statistical images.

A user first described this issue in 2019. While recent versions of dcm2niix prevent this unlikely situation, we should also make sure that AFNI correctly detects the error and terminates, rather than generating bogus images. This will protect users analyzing data converter with older versions of dcm2niix, using modern dcm2niix with different settings (to preserve disk space), or using different tools.