feat(event_queue): get_many outcome metrics + wider get latency buckets

[extreme4all]

What

• Widen event_queue_get_seconds histogram buckets from a 10s cap up to 30s (adds 15/20/30s, lower buckets untouched)
• Add event_queue_get_many_outcomes_total{queue, outcome} counter recording every successful get_many as full, partial, or empty
• Grafana queue dashboard: new stacked "get_many outcomes per second" panel

Why

• The latency histogram capped at 10s, so get_one idle waits (aiokafka getone() blocks until a record exists) all collapsed into the overflow bucket
• For get_many we want to know whether calls came back with the desired records or hit the consume timeout. len(result) < count exactly identifies timeout-driven returns in the kafka batcher loop (a full batch always reaches size >= count first), so no protocol or adapter changes were needed
• outcome distinguishes fully-empty polls from partial batches

Notes

• No config, interface, or adapter changes; memory backend empty polls count as empty outcomes
• kafka get_many can overshoot count across partitions, so >= count maps to full

Testing

• uv run mypy clean
• uv run ruff check / uv run ruff format --check . clean
• uv run pytest — 200 passed (new: parametrized _get_many_outcome classification + outcome/error metric tests)