Remove USE_COMPLEX_Z

[quantumsteve]

This removes the USE_COMPLEX_Z macro, building both double and std::complex<double> instantiations of each function. This also eliminates MYTYPE and MAGMA_T macros.

I also removed the USE_MALLOC macro and code contained as dmrg_malloc and dmrg_free do the same thing.

Lastly, I fixed some memory allocations preventing the batched plugin from working on multiple GPUs.

[quantumsteve]

@g1257 What in dmrgpp decides if we are using the real or complex code? single or double precision?

[g1257]

Complex is using only if either of the following conditions is met (1) usecomplex is in SolverOptions in the input, or (2) TimeStepTargeting is in SolverOptions See https://github.com/dmrgpp-project/dmrgpp/blob/master/dmrg/Engine/DmrgRunner.cpp#L70 If neither condition is met, then double is used.

[quantumsteve]

Complex is using only if either of the following conditions is met (1) usecomplex is in SolverOptions in the input, or (2) TimeStepTargeting is in SolverOptions See https://github.com/dmrgpp-project/dmrgpp/blob/master/dmrg/Engine/DmrgRunner.cpp#L70 If neither condition is met, then double is used.

OK, so instantiating both real and complex makes sense.

[g1257]

Yes, instantiating both real and complex makes sense.

[PDoakORNL]

This looks good to me but I would like to at least get a Magma CI environment ready and do a little testing on this. Whether the new CI flavor gets commit before or after is probably less crucial.