v2: EPSG:4326 to EPSG:2157 inconsistency with `proj` with negative latitude

[davidjb]

Many thanks for creating this library. I've tried the following transform between EPSG:4326 and EPSG:2157 and the calculations appear fine when inside the bounds of EPSG:2157. However, when outside the bounds and latitude becomes negative, there are in inconsistencies between wgs84 and other tools like proj. Testing something like Null Island (0,0) is fine and consistent, but once latitude is negative, results differ:

package main

import (
	"fmt"

	"github.com/wroge/wgs84/v2"
)

func main() {
	transform := wgs84.Transform(wgs84.EPSG(4326), wgs84.EPSG(2157)).Round(3)

	// Consistent
	east, north, h := transform(0, 0, 0)
	fmt.Printf("%.3f\t%.3f %.3f\n", east, north, h)

	// go run main.go
	// 1493322.696     -5179198.412 0.000
	// echo 0 0 | cs2cs +init=epsg:4326 +to +init=epsg:2157 -d 3
	// 1493322.696     -5179198.412 0.000
	// https://epsg.io/transform#s_srs=4326&t_srs=2157&x=0&y=-0
	// 1493322.696462611 -5179198.412407922

	// Inconsistent - negative lat
	east, north, h = transform(100, -1, 0)
	fmt.Printf("%.3f\t%.3f %.3f\n", east, north, h)

	// go run main.go
	// 12431088.352    -5548823.633 0.000
	// echo 100 -10 | cs2cs +init=epsg:4326 +to +init=epsg:2157 -d 3
	// 12431089.926    -24809902.916 0.000
	// https://epsg.io/transform#s_srs=4326&t_srs=2157&x=100.0000000&y=-10.0000000
	// 12431089.925768612 -24809902.916357957
}

Tested with the current commit on v2 at 5fb8989. Let me know if you need any more info.