tests: regression coverage for priority queue and ringbuf fixes

[gg582]

Summary

Adds regression tests for three recently fixed bugs. No library code changes.

1. Priority-0 task starvation/deadlock (fixed in 95854ca)

tests/test_sharded_sched.c — new test_pool_priority_zero_no_starvation:
submits 64 tasks all at priority 0 through the thread pool and asserts every
one completes via futures. The fixed bug encoded (priority 0, slot 0) to NULL,
silently dropping the task and deadlocking the submitter in ttak_future_get.
Also asserts the burst completes in under 10 seconds.

2. Ring buffer concurrent query data race (fixed in 439a7ff)

tests/test_ringbuf.c — new test_ringbuf_concurrent_query_race:
a producer and a consumer hammer push/pop while two querier threads
continuously call ttak_ringbuf_count/is_empty/is_full. Asserts the
per-snapshot capacity bound (count <= capacity) and sum sanity
(count == pushed - popped) after join, plus no crash. (Cross-function
agreements like count==0 == is_empty are intentionally not asserted: each
query takes its own lock, so they are separate snapshots of a concurrently
mutating buffer.)

3. Heap-backed priority queue restructure (cb804a9 / 95854ca)

tests/test_priority.c:

• test_priority_queue_zero_priority_slot0 — reproduces the slot-0 NULL
  encoding directly at queue level: a lone priority-0 task, and a priority-0
  task buried under 32 higher-priority pushes, must both survive push/pop.
• test_priority_queue_heap_stress — pushes 5000 items with a deterministic
  LCG (half priority 0, heavy duplication, occasional 42), pops all, and
  verifies pop order is monotonically non-increasing against a
  max-extraction model, with pop count exactly matching push count.

Test plan

• make test — full suite passes