remove variability on string generation

[mcollina]

Fixes #47

On node v14.5:

benchSonic*1000: 505.073ms
benchSonicSync*1000: 9.019s
benchSonic4k*1000: 471.747ms
benchSonicSync4k*1000: 338.145ms
benchCore*1000: 4.228s
benchConsole*1000: 7.986s
benchSonic*1000: 467.774ms
benchSonicSync*1000: 8.954s
benchSonic4k*1000: 463.897ms
benchSonicSync4k*1000: 326.755ms
benchCore*1000: 4.229s
benchConsole*1000: 7.845s

On node v14.4:

benchSonic*1000: 505.515ms
benchSonicSync*1000: 8.638s
benchSonic4k*1000: 455.108ms
benchSonicSync4k*1000: 326.902ms
benchCore*1000: 5.098s
benchConsole*1000: 7.555s
benchSonic*1000: 445.279ms
benchSonicSync*1000: 8.511s
benchSonic4k*1000: 446.532ms
benchSonicSync4k*1000: 303.18ms
benchCore*1000: 4.589s
benchConsole*1000: 7.377s

[mcollina]

@puzpuzpuz wdyt?

[puzpuzpuz]

Interesting. So, it looks like frequent string concats were creating GC pressure which was one of bottlenecks in the benchmark and GC algorithm started behaving differently in 8.3.

In that case it really makes sense to remove that bottleneck to focus the benchmark on the overhead of each stream implementation. The change makes sense to me.