Replace weighted semaphore with channel

[StephenButtolph]

Why this should be merged

Before:

Benchmark_Semaphore_Acquire-12                	79965123	        15.26 ns/op	       0 B/op	       0 allocs/op
Benchmark_Semaphore_Release-12                	77948438	        15.40 ns/op	       0 B/op	       0 allocs/op
Benchmark_Semaphore_TryAcquire_Success-12     	78838228	        15.01 ns/op	       0 B/op	       0 allocs/op
Benchmark_Semaphore_TryAcquire_Failure-12     	99081432	        11.77 ns/op	       0 B/op	       0 allocs/op

After:

Benchmark_Semaphore_Acquire-12               	56607668	        20.70  ns/op	       0 B/op	       0 allocs/op
Benchmark_Semaphore_Release-12               	55961013	        21.51  ns/op	       0 B/op	       0 allocs/op
Benchmark_Semaphore_TryAcquire_Success-12    	56662573	        21.54  ns/op	       0 B/op	       0 allocs/op
Benchmark_Semaphore_TryAcquire_Failure-12    	318408416	         3.779 ns/op	       0 B/op	       0 allocs/op

Because TryAcquire_Failure is the common case when hashing, we should optimize this path.

How this works

Replaces the semaphore package with a simple wrapper around a channel.

How this was tested

• New benchmark
• CI

[StephenButtolph]

We can wait until #2899 is merged to get a better e2e benchmark

[StephenButtolph]

(Running benchmarks on top of #2899, this does seem to be an improvement for larger tries when parallelism is enabled:

TL;DR: 5% improvement for large trie's. 10% regression for small trie's.

Before:

Benchmark_HashChangedNodes/1-12  		 3299204	       358.3 ns/op	     144 B/op	       3 allocs/op
Benchmark_HashChangedNodes/10-12 		   58080	     20863   ns/op	     721 B/op	      25 allocs/op
Benchmark_HashChangedNodes/100-12         	   10000	    114356   ns/op	    2872 B/op	     111 allocs/op
Benchmark_HashChangedNodes/1000-12        	    1384	    912711   ns/op	   17346 B/op	     813 allocs/op
Benchmark_HashChangedNodes/10000-12       	     152	   7474488   ns/op	   91583 B/op	    4331 allocs/op
Benchmark_HashChangedNodes/100000-12      	      14	  79664562   ns/op	 1156965 B/op	   65749 allocs/op

After:

Benchmark_HashChangedNodes/1-12  		 2710024	       400.2 ns/op	     144 B/op	       3 allocs/op
Benchmark_HashChangedNodes/10-12 		   52284	     22970   ns/op	     722 B/op	      25 allocs/op
Benchmark_HashChangedNodes/100-12         	    8569	    139937   ns/op	    2888 B/op	     111 allocs/op
Benchmark_HashChangedNodes/1000-12        	    1370	    926422   ns/op	   15252 B/op	     739 allocs/op
Benchmark_HashChangedNodes/10000-12       	     172	   7068256   ns/op	   89346 B/op	    4253 allocs/op
Benchmark_HashChangedNodes/100000-12      	      14	  75760193   ns/op	 1060367 B/op	   62297 allocs/op

It is a minor performance degradation for smaller tries as more of the TryAcquire calls succeed which is slower with this implementation.

[StephenButtolph]

This PR is being taken over by the PR to optimize key buffer reuse #2902