feat(queue): fast_forward_collaborative_group

[ltitanb]

Adds an opt-in way for a collaborative group to join at the producer's current write head instead of replaying whatever is still in the ring.

A fresh collaborative consumer's init_collaborative calls acquire_next_slot from a zero-initialised group cursor; the cursor immediately gets SpedPast and falls back to acquire_earliest_available_slot, landing at count - queue.len() — i.e. a full ring of backlog. That is correct for groups whose payload outlives the queue, but it's wrong when the payload references data with a tighter lifetime (e.g. an external arena that recycles independently). Replaying the backlog there produces stale references and noisy warnings.

Queue::fast_forward_collaborative_group(label) advances the group cursor via fetch_max(self.count()), allocating the slot if needed. Exposed on FluxSpine for any field with AsRef<SpineQueue<T>>.

Safety: must be called once, before any group member starts consuming. After a member's acquire_next_slot has run, fetch_max can overshoot already-claimed positions and silently drop messages produced in the gap. Documented on the API.

Tests: single-consumer skips backlog; 4-consumer group sees every post-fast-forward message exactly once.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 3 additional findings.