cmd/compile,runtime: implement uint64->float32 correctly on 32-bit archs

cmd/compile,runtime: implement uint64->float32 correctly on 32-bit archs

The old way of implementing it, float32(float64(x)), involves 2 roundings
which can cause accuracy errors in some strange cases. Implement a runtime
version of [u]int64tofloat32 which only does one rounding.

Fixes #48807

Change-Id: Ie580be480bee4f3a479e58ef8dce23032f231704
Reviewed-on: https://go-review.googlesource.com/c/go/+/354429
Trust: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>