When handling the `negate` ZIR instruction, Zig now checks for a
comptime operand and handles it as a special case rather than lowering
it as `0 - x` so that the expression `-x` where `x` is a floating point
value known at compile-time, will get the negative zero bitwise
representation.