Faster filtering mapping

[jeswr]

Supercedes #48
Supercedes #50

Makes use of this suggestion by @RubenVerborgh to provide the ~2x speedup to chained maps and transforms; without having the problems associated with immutability (to give you a sense, the test suite is now running at ~600ms on my machine vs ~1000ms before).

[jeswr]

The performance improvements greatly depend on the number of filters. In the example @jacoscaz had with only maps - there was no gain (if anything a slight degradation of 5-10%), but if there are filters involved the speedup can be much more significant. For instance running

range(0, 2_000_000)
.map((item: number) => item)
.filter((item: number) => item % 2 === 0)
.map((item: number) => item)
.filter((item: number) => item % 3 === 0)
.map((item: number) => item)
.filter((item: number) => item % 4 === 0)
.map((item: number) => item)
.filter((item: number) => item % 5 === 0)
.map((item: number) => item)
.filter((item: number) => item % 6 === 0)

Goes from ~80ms to ~20ms; so a 4x speedup

As a result - it may be worth only triggering this optimized iterator when we reach the first FilterIterator which we can do by simply removing the following lines https://github.com/jeswr/AsyncIterator/blob/1ec39fbb12497ad46ea5409ba683deb4cbb2079c/asynciterator.ts#L1331-L1351

[jeswr]

I'm not sure I understand the cause of the failing test on node v10

[jeswr]

As discussed on gitter @jacoscaz will merge this into #48 - I will close this once that has been done.

[jacoscaz]

Done!