Testing both single and double precision together can hide issues

As a follow up to PR #445 , we should change our CACTS configuration to test single precision _separately_ from double. The reason is that if `EKAT_TEST_DOUBLE_PRECISION` is defined, our unit tests use `Real=double`, regardless of anything else. In order to force a test where `Real=float`, we need to have only one between `EKAT_TEST_DOUBLE_PRECISION` and `EKAT_TEST_SINGLE_PRECISION` defined.

My reccomendation would be to test both sp and dp in debug builds, and only dp in release builds (which matches what we already do in eamxx anyways).