perf: use a sync pool for machine constructor

[peter7891]

This PR uses a sync.Pool in the Machine constructor.
This improves the following

1. Reduces allocations and deallocations
2. Improves runtime and memory consumption
3. Because of less objects in memory, less GC pauses

issue with benchmarks from master.

Benchmarks on this branch.

CPU bench

Type: cpu
      10ms   100%   100%       10ms   100%  github.com/gnolang/gno/pkgs/gnolang.(*Machine).PopAsPointer
         0     0%   100%       10ms   100%  github.com/gnolang/gno/pkgs/gnolang.(*Machine).Run
         0     0%   100%       10ms   100%  github.com/gnolang/gno/pkgs/gnolang.(*Machine).RunMain
         0     0%   100%       10ms   100%  github.com/gnolang/gno/pkgs/gnolang.(*Machine).RunStatement
         0     0%   100%       10ms   100%  github.com/gnolang/gno/pkgs/gnolang.(*Machine).doOpAddAssign
         0     0%   100%       10ms   100%  github.com/gnolang/gno/pkgs/gnolang.BenchmarkBlah
         0     0%   100%       10ms   100%  testing.(*B).launch
         0     0%   100%       10ms   100%  testing.(*B).runN

Memory profile

Showing nodes accounting for 10958.29kB, 72.79% of 15054.80kB total
Showing top 10 nodes out of 112
      flat  flat%   sum%        cum   cum%
 4097.83kB 27.22% 27.22%  4097.83kB 27.22%  github.com/gnolang/gno/pkgs/gnolang.addAssign
 1537.50kB 10.21% 37.43%  1537.50kB 10.21%  runtime.allocm
 1184.27kB  7.87% 45.30%  1184.27kB  7.87%  runtime/pprof.StartCPUProfile
  902.59kB  6.00% 51.29%  1553.21kB 10.32%  compress/flate.NewWriter
  650.62kB  4.32% 55.62%   650.62kB  4.32%  compress/flate.(*compressor).init
  528.17kB  3.51% 59.12%   528.17kB  3.51%  regexp.(*bitState).reset
  519.03kB  3.45% 62.57%   519.03kB  3.45%  time.init
  513.50kB  3.41% 65.98%   513.50kB  3.41%  github.com/cockroachdb/apd.(*Decimal).setString
  512.50kB  3.40% 69.39%   512.50kB  3.40%  github.com/gnolang/gno/pkgs/amino.(*Codec).parseStructInfoWLocked
  512.28kB  3.40% 72.79%  1024.29kB  6.80%  github.com/gnolang/gno/pkgs/gnolang.(*defaultStore).SetCacheType

As we can see, we no longer have the Machine constructor in the top 10 of either.
And we can see that the memory consumption went from 28666.37kB to 15054.80kB.

[jaekwon]

Nice. What was used for benchmarking?

Added one comment for feedback above.

[peter7891]

Nice. What was used for benchmarking?

Added one comment for feedback above.

I ran the gno code like this, in the benchmark.

	n := MustParseFile("main.go", c)
	m.RunFiles(n)
	m.RunMain()

And for the profiling, i used.

go tool pprof

[moul]

@jaekwon, do you think @peter7891's answer is satisfactory? Can we merge his work? Should we add any comments or issues for future improvements, or include unit tests?

[peter7891]

@jaekwon, do you think @peter7891's answer is satisfactory? Can we merge his work? Should we add any comments or issues for future improvements, or include unit tests?

The only thing i could possibly see adding a test for is to make sure the values from a fetched object are zero-ed.
I am not sure how valuable that is though. If that wasn't the case, all other functionality would break and thus other tests would break.
We can't detect if someone didn't release the object, which was @jaekwon concern.
So...

[jaekwon]

LGTM