feat: Add flatmapConcat with parallelism support

[He-Pin]

Motivation:
Add support for #1672
Based on: #1701

Modification:

1. Add flatmapConcat with parallelism
2. add optimization to avoid materialization at the same time.

Result:

1. flatmapConcat with parallelism supported

Benchmark:

[info] Benchmark                                       Mode  Cnt         Score         Error  Units
[info] FlatMapConcatBenchmark.completedFuture         thrpt    5   1045035.962 ±   95438.439  ops/s
[info] FlatMapConcatBenchmark.completedFutureP1       thrpt    5  10200822.173 ±  506488.622  ops/s
[info] FlatMapConcatBenchmark.internalSingleSource    thrpt    5  17561424.472 ± 2469846.548  ops/s
[info] FlatMapConcatBenchmark.internalSingleSourceP1  thrpt    5  17434317.937 ± 2230811.585  ops/s
[info] FlatMapConcatBenchmark.mapBaseline             thrpt    5  41100524.324 ± 5565726.687  ops/s
[info] FlatMapConcatBenchmark.normalFuture            thrpt    5    312508.444 ±   45271.476  ops/s
[info] FlatMapConcatBenchmark.normalFutureP1          thrpt    5    327743.914 ±   29201.871  ops/s
[info] FlatMapConcatBenchmark.oneElementList          thrpt    5   1162074.042 ±   78000.108  ops/s
[info] FlatMapConcatBenchmark.oneElementListP1        thrpt    5  10988702.693 ±  812309.621  ops/s
[info] FlatMapConcatBenchmark.sourceDotSingle         thrpt    5  11460931.634 ± 1107362.582  ops/s
[info] FlatMapConcatBenchmark.sourceDotSingleP1       thrpt    5  11313667.518 ± 1521344.870  ops/s

jmh:run -i 5 -wi 5 -f1 -t1  org.apache.pekko.stream.FlatMapConcatBenchmark

[He-Pin]

I hate 2.12.x

[jrudolph]

I wonder if it is really necessary, wouldn't it be easier to just prematerialize to get the same effect?

[He-Pin]

Yes, but I think it would be nice to have this out of box. otherwise will need a mapAsync and prematerialize and then flatmapConcat, and I think the current implementation will have better performance than that too.