[feat] PIP-468: V5 flow-control knob test coverage

[merlimat]

Summary

Coverage gap from the V5 audit — the plumbing for these knobs was wired but never exercised in any test, so a regression that broke the V5 → v4 mapping would have shipped silently. Pins the contract for four flow-control settings.

Test class	Tests	What it covers
V5ConsumerPriorityLevelTest	2	Behavioral: high + low priority consumers on a Shared subscription. With prefetch drained by acks, the high-priority consumer receives every message; with prefetch full (no acks), the broker overflows to the low one. Single-segment topic to keep per-segment dispatch deterministic.
V5ConsumerReceiverQueueSizeTest	2	receiverQueueSize(int) propagates from the V5 builder to every per-segment v4 ConsumerImpl — verified via reflection on segmentConsumers + ConsumerBase.getCurrentReceiverQueueSize(). Covers single-segment and 3-segment topics.
V5ProducerFlowControlTest	4	sendTimeout(Duration) and blockIfQueueFull(boolean) propagate from the V5 builder to every per-segment v4 ProducerImpl — verified via reflection on segmentProducers + ProducerBase#conf. Tests cover blockIfQueueFull(true), blockIfQueueFull(false), single-segment sendTimeout, and 3-segment sendTimeout. Sends warm-up messages before reading state because V5 segment producers are created lazily on first send.

Why config-propagation tests for the producer knobs and not behavioral? Behavioral verification of the actual timeout-firing and queue-full paths lives in the v4 test suite (SimpleProducerConsumerTest.testSendTimeout etc.); those tests stop the broker mid-send to force the pending-queue overflow / timeout, which the in-process SharedPulsarBaseTest cluster used here cannot do. The plumbing tests added here suffice as a regression guard for the V5 → v4 mapping.

Test plan

• All 8 new tests pass (8 of 8).
• Full V5 broker test suite green (133/133).
• pulsar-broker checkstyle clean.