Refactor and fix getBulk retry logic

[shy-1234]

Before this change

• When getBulk had a complete miss on the first attempt, the retry logic would stop retrying as long as at least one result was returned.
• When getBulk had a partial miss on the first attempt, chunk read timeouts were never reflected in the metrics because hasZF was always true (unlike the complete-miss retry logic, where hasZF is set to false when it reaches the very last client).

After this change

• getBulk complete miss and partial miss now share the same retry logic. The client will continue retrying to fetch missing keys, if any, until it has exhausted all available replicas.
• When getBulk has a partial miss on the first attempt, it will now set hasZF to false when it reaches the last client, and will therefore correctly report chunk read timeouts to metrics when they occur on the last attempt (which is the one that should be reported).