Add Linux IO::Event::Futex support

[samuel-williams-shopify]

Description

Add a Linux-only IO::Event::Futex backed by an aligned 32-bit word in an IO::Buffer. The platform primitive lives in futex.c/futex.h; uring.c contains only the optional io_uring integration.

The futex provides atomic value access, increment-and-wake signalling, single waits, and vector waits over up to IO::Event::Futex::WAITV_LIMIT futex words. Futex#wait returns whether it slept, while Futex.wait_any returns the index of the futex that woke or nil when an expected value had already changed.

When a fiber scheduler implements futex_wait or futex_waitv, the native methods delegate to it. The URing selector provides cancellable asynchronous implementations using IORING_OP_FUTEX_WAIT and IORING_OP_FUTEX_WAITV. Without a scheduler, the same native methods call the Linux futex syscalls while releasing the GVL. A scheduler that does not expose the corresponding optional hook raises NotImplementedError rather than blocking its reactor thread.

The class is defined only on Linux and does not require io_uring. The asynchronous selector hooks are exposed separately when io-event was compiled with a sufficiently recent liburing and the running kernel reports support for the corresponding opcodes.

Motivation

This provides a low-overhead shared-memory notification path for processes that already exchange state through IO::Buffer. Vector waits allow a coordinator such as Fantail to sleep until any worker permit changes without polling every worker. Consumers can capability-detect IO::Event::Futex and use a portable notification mechanism when it is unavailable.

Testing

• Full test suite on macOS: 281 passed, 9 skipped.
• RuboCop passes for all changed Ruby files.
• Documentation coverage remains 100%: 93/93 definitions.
• Dedicated Ubuntu CI builds liburing 2.6 and requires both Futex.wait_any and selector futex_waitv support.
• Linux tests exercise synchronous no-scheduler waits, scheduler-driven io_uring waits, value mismatches, independent futex words, the exported wait-vector limit, and vector wake indices.

The local Colima VM does not permit io_uring setup under its seccomp policy, so the native asynchronous runtime path is covered by the dedicated Linux CI job.