F32.round loses the sign of negative zero #1611

F32.round turns negative zero into positive zero. It does the same when a negative fraction rounds to zero, including the negative half under the existing ties-toward-positive-infinity rule.

import Base

def main() -> IO(Unit):
  do IO<Unit>:
    IO.print(U32.show(F32.bits(F32.round(F32.neg(0.0)))))
    IO.print(F32.show(F32.div(1.0, F32.round(F32.neg(0.1)))))

On 7e120d5, this prints 0 and inf; preserving the sign would give 2147483648 and -inf. Current main has the same definition. The sign is observable through both the public bit conversion and ordinary arithmetic.

I kept the existing tie rule and restored the input sign only when the rounded result is zero. Negative zero, negative fractions, the half boundary and the number kit now pass on Bun and native C. AI-assisted investigation and fix.