[NPM/YARN] Cache client-side timeouts when a remote host is unreachable

[brrygrdn]

Follows up on #5142

This PR modifies the timeout caching strategy we used for Maven into a generic 'Dependabot::RegistryClient' and then applies it to the npm_and_yarn gem to ship this strategy for another ecosystem.

NPM is currently the ecosystem experiencing this problem most often, so let's address it next.

Alternatives considered

I originally approached this as an Excon middleware, but there were a few issues with that approach in my opinion:

• I'd prefer to roll out the change gradually one ecosystem at a time to avoid missing any unexpected interactions with package managers
• I'd prefer to limit the change in Excon behaviour to code that is calling registries rather than all remote calls.
  • We could do this by parameterising the SharedHelpers to incorporate the middleware for some callers but not others, but that feels somewhat brittle
• Since we are now caching state I'd prefer to do it within a concrete class, a middleware could be made to call out to a data store class but the lifecycle is slightly harder to reason about when it is part of an Excon stack

[jurre]

This approach makes sense to me 👍