RFE: Add limit caching

[Lance Bragstad]

The oslo.limit library is expected to make calls to keystone on behalf of the service using it to fetch limits. To prevent extra network chatter, we should offer the ability to cache these responses and make the TTL configurable for deployers. If deployers want the most accurate limit information from keystone, they should set a low TTL or disable caching, which is a sane default behavior. If a deployment is concerned about chatty networks when limit information the deployment is static, they can configure caching to opt into better performance.