Adopt glTF convention: lights face local neg-z direction

[WestLangley]

As discussed in #14775 (comment).

For now, only RectAreaLight is governed by its quaternion, so it is the only light for which this new convention applies.

This change required a rewrite of RectAreaLightHelper. The helper must now be added as a child of the light, which is probably a better API anyway.

[WestLangley]

/ping @donmccurdy

[donmccurdy]

glTF does not currently define area lights, so this isn't a conformance concern. However — do we think similar changes to other directed lights (Spot, Directional) will eventually be wanted? I.e. removing light.target in favor of local direction affected by the node's world rotation? If that's the end goal, then +1 for updating RectAreaLight and using -Z for lights going forward. If we think the other directed lights will remain as-is, then I have no preference on this PR.

[WestLangley]

However — do we think similar changes to other directed lights (Spot, Directional) will eventually be wanted? I.e. removing light.target in favor of local direction affected by the node's world rotation?

I believe so.

[mbredif]

+1 for having the same convention for lights and cameras (lookat faces neg-z instead of pos-z for lights and neg-z for cameras), as it simplifies shadowmapping with this shared convention (proposed in #14658 (comment) already).

[mrdoob]

This change is scary, but less see what happens...

[WestLangley]

We'll see... Thanks for giving this PR a go!

[mrdoob]

Thanks!