Fix instances

[Jaxan]

Should fix issue #41

There were a couple of bugs in the old (commented out) instances:

• Scalar (Complex t) used to be Scalar t but should really be complex
• norm2Sq for basic types was implemented as abs and not as **2
• normalize _ always returned 1 for basic types, but should return -1 for negative values

I tried to keep instances (and their constraints) simple. But I think that Normed (SpVector t) could potentially be simplified. Also the I changed one constraint for the LinearVectorSpace class, because that was a bit easier.

I re-enabled many tests. All but 1 succeed.

[ocramz]

@Jaxan the only test that's currently failing is the complex-valued Cholesky factorization. This has really bugged me for a while, though your PR is unrelated to this. I'll try to produce a minimally-breaking example.