stream: improve read() performance

[mscdex]

Benchmark results:

                                                        confidence improvement accuracy (*)    (**)   (***)
 streams/readable-bigread.js n=1000                           ***      9.17 %       ±0.96%  ±1.28%  ±1.66%
 streams/readable-unevenread.js n=1000                        ***      5.02 %       ±1.06%  ±1.41%  ±1.84%

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/25259/

[Fishrock123]

This code is frankly quite difficult to understand, but I think this is correct.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/25294/

[Trott]

@nodejs/streams

[mcollina]

LGTM, good work! A CITGM run would be helpful to assess any breakage.

[mscdex]

@mcollina What breakage? These changes consist of:

• Replacing src.copy() with dest.set() (like was done in one of my recent streams perf PRs)

• Folding the copy of the first list item into the loop

• Avoiding unnecessary calculations once we've reached the last buffer in the list used to fill ret

[mcollina]

I'm skeptical of any side effect a change in streams can have at this point, so a run would not be a waste of time.

CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1987/

[mcollina]

I don't see anything suspicious on CITGM, LGTM from me.