According to many sources I've looked through, part of the definition of trace-class operators is that they be compact. What's the need for this caveat? Why not just look at all those operators $T$ for which $\sum_{i \in I} \langle T e_i,e_i \rangle$ converges absolutely for some orthonormal basis $\{e_i\}_{i \in I}$? This is the definition of trace-operator with the compactness assumption dropped, right?
As an aside, is there an example of an operator $T$ that is not compact yet $\sum_{i \in I} \langle T e_i,e_i \rangle$ converges absolutely for some orthonormal basis $\{e_i\}_{i \in I}$