read path: slot based job scheduling

[mapno]

What this PR does:

Query latency is driven by the slowest job, and today's batch protocol adds another barrier: completed jobs are not returned until every job in the batch finishes. This proposal makes the job scheduling unit while retaining coalesced transport.

Jobs become the scheduling unit, consuming individual slot in queriers. However, for efficiency, jobs are still transported in batched called frames.

Each querier has a configured number of execution slots, with completed jobs returning results and freeing their slots for new work without waiting for the slowest job in their original batch. This reduces idle capacity and prevent a slow job from delaying completed results from other queries. Capacity and dispatch remain explicit and easy to monitor; running stragglers stay assigned, so the improvement comes from reusing freed slots while work is still queued.

Checklist

• Tests updated
• Documentation added
• Changelog entry added under .chloggen/ (run make chlog-new, or make chlog-new FILENAME=<name> to override the default branch-name file; see .chloggen/README.md)