add support for two accelerators

[psychocoderHPC]

• add trait IsThreadSeqAcc
• change internal methods to queue a kernel
• add AccThreadSeq accelerator

more details:

• if one accelerator is enabled cupla::AccThreadSeq is equal to cupla::Acc
• This pull request allow under special conditions to enable two accelerators. This double selection is only useful for porting CUDA code to cupla. conditions:
  • one accelerator must be work fully sequential in the thread level
  • the other must be support a arbitrary number of threads in the thread level
• CUPLA_KERNEL_OPTI use always cupla::AccThreadSeq as accelerator