blk-mq: prefer current CPU for async hctx run work - #12

Summary

Prefer the current CPU for blk-mq async hardware-queue run work when that CPU is already part of the hctx CPU mask.

Existing unbound and round-robin fallbacks remain in place when the current CPU is not suitable.

Why

The previous logic always rotated through the hctx CPU mask in batches. If the caller is already on a valid CPU, scheduling work elsewhere can create avoidable remote wakeups and cacheline movement.

Compatibility

No userspace ABI, module contract, exported symbol, Kconfig, or dependency changes. This only refines internal CPU selection for async hctx run work.

Validation

• Passed targeted ARCH=um object build for block/blk-mq.o.
• x86 object validation was attempted locally but blocked before compiling this object because the host environment lacked gelf.h, required by objtool.