OpenAI has removed the ability for keys to query for their own usage, effectively killing usage tracking. We can still get keys' limits but we don't know how much of that limit has been used.
This reworks the keychecker and keypool to remove all usage checking/reporting code and adjusts the checker to now test all keys for liveness. The key check interval has also been increased from 5 minutes to 30, since there's not much of a reason to check keys frequently now that we can't see their usage. Accordingly, we also test every key for liveness during each check in case the proxy is infrequently used or has many keys in its pool.
We can test keys without actually spending any quota by requesting max_tokens: -1 and checking for a specific Bad Request error message (instead of a 429, which indicates the key may be out of quota or dead).
On the bright side, key checking is much faster now.
