When adding the initial count for a rate limit window in cache, the

expiration was being set based on the window value, which was too large

to be accepted by the cache backend, being a UNIX timestamp and not a

number of seconds in the future. This resulted in the add failing.
We now use the number of seconds remaining + 60 instead, which isn't

exactly the same but gives us a suitable expiration with a buffer built

in.

Tested this build in production and verified it fixed the rate limit

caching.