Adds an optional settings to use Yarn to run Husky

[arcanis]

This fix has multiple purposes that improve the compatibility with Yarn at little cost:

• First, it ensures that the Husky command will run with the same pipeline as the one used by Husky's users when they run their scripts. It previously wasn't the case, because Husky was using run-node to "detect" the version of Node to use whereas Yarn users were using yarn run [...] (which enforces the Node version to be the same one as the one used to run Yarn itself).

• Second, starting from Yarn 2 we're looking to keep packages inside their tarballs as much as possible. In the context of run-node this causes problems, because run-node is a shellscript and can only be executed when unpacked. Using yarn husky-run avoids the issue altogether.

• Third, when working with PnP environments, if we call the Node script manually, then we need to inject the PnP hook so that Husky knows how to load its dependencies (since they aren't in the node_modules folder anymore). Using yarn run makes this process transparent: it will inject the hook when running under PnP, and won't do it otherwise, and the Husky script won't have to know anything about it.

• Four, I'd like to make it possible to track which scripts and binary are used in an application (through Yarn's new plugin system). Directly calling the Husky script wouldn't allow that, but going through yarn run would.

I guess later we can see whether it would make sense to enable HUSKY_USE_YARN by default in some circumstances, but for now I think it's fine if we keep it behind a variable that must be set through a repository-local rcfile (cf #510).