The PALI rate limiter currently lets requests larger than the base bucket size bypass rate limiting entirely (due to folly token bucket implementation) and this can bypass our configured rates

While investigating of HELP-101050 we saw multiple large page writes and append timeouts. Bypassing PALI rate limiting ended up in us producing extra traffic sent to LS and resulted in us being uxcessively rate limited by LS rate limiter

Requests larger than our base bucket size should still be rate limited to respect our rate. We must also ensure that large requests eventually make progress and don't get stuck forever.