Bug 3758 – Create D impementation of to!(float, string), etc.

std.conv calls the c functions atof, atod, atold, etc. for to!(float, string), to!(double, string), and to!(real, string) respectively. There are two practical reasons why we need D implementations of these functions: 1. Strings should be convertible to floats at compile time. 2. To make the input string compatible with C, it is copied using toStringz. This is inefficient, especially when trying to real large files with ASCII numbers in and convert the ASCII numbers to floats.

A D implementation that doesn't allocate on every conversion now exists, even though it still doesn't work at compile time. Since converting strings to floats at compile time isn't very useful in practice anyhow, and I cared more about the memory allocations when I initially filed this, I'll mark it as resolved.