perf(semaphore): Allocate wake buffer only when needed

[mhambre]

Avoids unnecessary wake buffer allocation on the semaphore release hot path when no tasks need to be woken.

cargo bench -p benchmarks --bench benchmarks -- semaphore mutex \
  --min-time 0.75 \
  --sample-count 100 \
  --color never

Median

Benchmark	Before	After	Improvement
Mutex uncontended reuse	57.94 ns	21.15 ns	63.5%
Semaphore try-acquire/release	51.10 ns	13.82 ns	73.0%
Semaphore owned acquire/release	52.40 ns	22.61 ns	56.9%
Semaphore release	90.83 ns	43.93 ns	51.6%
Repeated permit debt	4.124 µs	1.728 µs	58.1%

Mean

Benchmark	Before	After	Improvement
Mutex uncontended reuse	57.99 ns	21.24 ns	63.4%
Semaphore try-acquire/release	51.00 ns	13.85 ns	72.8%
Semaphore owned acquire/release	54.46 ns	22.62 ns	58.5%
Semaphore release	94.47 ns	43.62 ns	53.8%
Repeated permit debt	4.124 µs	1.731 µs	58.0%

[tisonkun]

Thanks!