Optimise message expiry

[kjnilsson]

Optimise message expiry
Problem
The expire function in rabbit_fifo processed expired messages one at a time in a recursive loop: it would take a single message, call discard_or_dead_letter for that one message, update the state, and then recurse via expire_msgs to check for the next expired message. This meant that if N messages were expired, discard_or_dead_letter was called N times (once per message), and the state record was rebuilt N times.

Similarly, expire_shallow (used by the expire_msgs timer callback) only expired a single returned message from the returns queue before moving on to the messages queue.

Changes
Replace the recursive one-at-a-time expire/expire_msgs loop with a batch-oriented expire_batch that collects all expired messages from both the returns queue and the messages priority queue in one pass, then processes them with a single discard_or_dead_letter call
Extract take_expired_returns/2 to drain all consecutive expired messages from the returns queue (previously only the head was checked)
Extract count_and_size/1 to compute the count and total byte size of expired messages in a single fold
Reuse expire_batch from expire_shallow instead of duplicating the expiry logic
Effect
Reduces the number of discard_or_dead_letter calls from N (one per expired message) to 1 per expiry pass
Reduces intermediate state record allocations from N to 1
Eliminates code duplication between expire_shallow and expire