feat: replace q.q_ring with thread-local variable and introduce 3 APIs

Replace individual q.q_ring usage with a unified thread-local QUEUE_RING
using OnceCell for lazy initialization. This change provides cleaner API
access through helper macros and improves thread safety.

Key changes:
- Add thread-local QUEUE_RING with OnceCell<RefCell<IoUring>>
- Create with_queue_ring() and with_queue_ring_mut() for easier external access
- Replace all q.q_ring.borrow() and q.q_ring.borrow_mut() calls (20+ instances)
- Update both src/io.rs and src/uring_async.rs
- Add proper type annotations for compile-time safety
- Initialize thread-local ring when queue is created

Benefits:
- Cleaner, more ergonomic API through helper macros
- Thread-local access avoids cross-thread synchronization
- Lazy initialization reduces memory overhead
- Type-safe access with explicit annotations

Three new APIs are added:

- with_queue_ring()

- with_queue_ring_mut()

- init_task_ring()

The 1st two are for manipulating the queue's ring, and the 3rd one can
be used for customerizing the task ring.

Per-task io_uring is the most common usage, so take this way for
libublk.

- helps to use task ring to handle multiple queues from different
  devices

- helps to support async/.await control command because control ring
  actually depends on io ring if the two are in same task context

- with control command async/.await ready, it will be pretty
  easier/simpler to run arbitrary queues & tasks & devices combination,
  one typical example is to create many devices in single process, and
  let one thread to handle queues with same index from different
  devices.

Signed-off-by: Ming Lei <tom.leiming@gmail.com>

Author: ming1