CTS tests for `length` and `distance` expect `abs`-based calculation for 1D, but spec is ambiguous

[andyleiserson]

The CTS tests for length and distance expect an abs-based calculation to be used when the argument is 1D (i.e. a scalar). This avoids overflows that are possible in the $\sqrt{\sum e_i^2}$ calculation.

However, the spec is a bit confusing on this point. It does say that length:

evaluates to the absolute value of e if T is scalar.

But it also says in a note that:

the scalar case may be evaluated as sqrt(e * e), which may unnecessarily overflow or lose accuracy.

The floating-point accuracy table says of length:

Inherited from sqrt(dot(x, x)) in the vector case, and sqrt(x*x) in the scalar case.

[andyleiserson]

After revising the description I'm less sure that this belongs here vs. as a spec issue.