perf(xid_correlation): lower allocs/event from 12.4 toward NORTHSTAR (2)

[trilamsr]

Tracks per-detector alloc optimization toward NORTHSTAR (2 allocs/event) for
xid_correlation. Current ceiling in scripts/bench-registry.sh::allocs_gate
is 13 allocs/event (measured 12.40 ev, 12699 allocs/op @ window=1024).
The target is ≤2 to converge on NORTHSTAR alongside the other detectors.

What the bench measures

./bench/detectors/BenchmarkXidCorrelationDetector — 64 Xid records × 1024
pod-eviction events, joined by node + time. One verdict per evicted pod
(most-recent Xid wins). See bench/detectors/detectors_bench_test.go.

Hot-path candidates (pre-investigation)

• One verdict allocated per evicted pod (~819 pods); verdict structs +
  evidence trail dominate the allocation count.
• Per-pod string formatting for evidence strings.
• ObjectRef Pod-kind allocation per event.

Definition of done

• Lower the ceiling in scripts/bench-registry.sh::allocs_gate to the new
  floor (the absolute gate prevents creep back upward).
• Update bench/detectors/baselines.json for the soft-gate trend tracking.
• Profile-driven (go test -bench=BenchmarkXidCorrelationDetector -memprofile)
  — fix the largest contributor, re-measure, repeat.

Why now (priority signal)

Above NORTHSTAR by ~6×. Highest leverage of the remaining gap (along with
pod_evicted, which is tracked separately).

Rolls up to #302.