Only warn once when a flag is missing

[erbridge]

It's totally legitimate to reference a feature flag that doesn't exist, yet, but before this change, doing so would spam our logs with warnings.

[philandstuff]

Hmm, I think it might be a useful signal that a flag has been typoed? It does mean we need to remember to create the flag before deploying though, otherwise we get said spam.

[erbridge]

Has that ever happened for us? Because I know we do get it right, but don't create the flag in LaunchDarkly, resulting in said spam.

[nickstenning]

IMO most ergonomic thing to do here would be to keep a package private map[string]bool which says whether or not we've emitted the warning, and emit it once before setting the key value to true in that map.

[erbridge]

Nice. I like that idea.

[philandstuff]

as long as said map is thread-safe. A sync.Map, say.