Add automatic batch deprioritization for large submissions

[helq]

On top of #415

We receive large requests into our sashiko instance (upon thousands of patchset reviews in a single minute). To keep sashiko servicing reviews that don't come in the batch, this change lowers the priority of patchsets that arrive within the same window of time (30 seconds). How many requests and how large the window should be completely depend on our use case. We use:

        # Bulk deprioritization: cluster same-repo submissions within a 30s
        # window; <10 unchanged, 10-39 -> 400, >=40 -> 300.
        - name: SASHIKO__REVIEW__BATCH_WINDOW_SECS
          value: "30"
        - name: SASHIKO__REVIEW__BATCH_TIERS__0__MIN_SIZE
          value: "10"
        - name: SASHIKO__REVIEW__BATCH_TIERS__0__PRIORITY
          value: "400"
        - name: SASHIKO__REVIEW__BATCH_TIERS__1__MIN_SIZE
          value: "40"
        - name: SASHIKO__REVIEW__BATCH_TIERS__1__PRIORITY
          value: "300"

---

When a patchset transitions to Pending, count how many sibling
patchsets share a similar timestamp (within a configurable window,
default 30s). If the count exceeds configured thresholds, cap
the priority of all patchsets in that window using MIN(priority, cap)
to avoid blocking the review queue with massive batch syncs.

Three new config options in [review]:

• batch_window_secs: time window for sibling detection (default 30)
• batch_tiers: list of {min_size, priority} threshold tiers

The deprioritization is applied retroactively to all patchsets in
the window on each new arrival, ensuring early arrivals in a batch
get deprioritized once the batch size becomes apparent. The MIN()
semantics ensure priority is never raised, preserving any manual
or rule-based deprioritization already in place.