fix(convertPathData): use relSubpoint, not .coords

[KTibow]

it turns out in the past i used item.coords (post-instruction coordinate in ORIGINAL svg) (resulting in incorrect calculations) when i should've been using relSubpoint; this switches to relSubpoint when fitting.

closes #2093

[KTibow]

the failing test cases are things like

<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 200 200">
<path d="M 78.301922,39.376734 L 64.971121,61.742246 L 73.061784,64.348229 L 78.512656,70.6691 L 84.285326,65.033743 L 92.451569,61.727261 L 78.301922,39.376734 z" fill="none" stroke="black" stroke-width="2"/>
</svg>

and the fail is because SVGO incorrectly rounds in a way that makes the start and end no longer connected

and the incorrectly coded "convert going home to z" logic was the only thing stopping the failure from surfacing

so fixing one thing shows another broken thing causes regressions... oh well i'll make a fix