wasm: replace queued locks with atomic ticket locks - #45

The generic queued spinlock relies on coherent mixed-width atomic accesses. WebAssembly only orders atomic accesses to identical byte ranges, while the current unlock and slow paths emit plain byte and halfword stores overlapping 32-bit atomic RMWs.

Use architecture-owned FIFO ticket spinlocks and writer-ticket rwlocks instead. Each concurrently accessed field occupies its own u32 and every access uses a WebAssembly atomic operation. Contended waiters sleep on the changing field and are notified directly on handoff.

This removes the unsupported mixed-width access pattern while preserving FIFO progress for spinlock and writer acquisition.

Validation:
- full Wasm kernel build
- wasm-objdump confirms lock state uses only i32.atomic load/RMW, memory.atomic.wait32, and memory.atomic.notify
- temporary four-CPU validation passed 40,000 contended spin acquisitions and 20,000 writer acquisitions with exact counters and no ownership or reader/writer overlap errors
- five-run lifecycle comparison showed no material boot-time regression (5.22 s mean versus 5.32 s for successful qspin baseline runs)

The intermittent workqueue RuntimeError is not claimed fixed here: one of 30 lock-only lifecycle attempts hit an uncaptured RuntimeError, while 20 immediately repeated sequential attempts were clean.

Agent-Session: codex:019fe115-244a-76c1-b716-b8af6827da00