lib: use utf8 fast path for streaming TextDecoder

[ChALkeR]

Tracking: #61041

A continuation of #61409

This is based on the logic in https://github.com/ExodusOSS/bytes

Unifies stream and non-stream codepath for UTF-8 both for intl and no-intl variants.

Previously, ICU and string_decoder were used on with-intl and without-intl variants correspondingly for streaming UTF-8 implementation

Instead, just do minor quick slices on JS side and use the single stateless native decoding API to be fast on large chunks

This:

1. Improves streaming UTF-8 TextDecoder performance
2. Brings streaming fatal UTF-8 TextDecoder support to without-intl builds
3. Unifies the logic, using a single codepath everywhere for UTF-8

Benchmarks

Previously: #61131

Non-chunked, 25.5.0

Test	Size	Throughput	Mean Time
Latin lipsum (ASCII)	84.902 KiB	36.69 GiB/s	0.002 ms
Arabic lipsum	79.771 KiB	2.03 GiB/s	0.037 ms
Chinese lipsum	68.203 KiB	4.10 GiB/s	0.016 ms
Arabic + 2 * ASCII	249.577 KiB	2.72 GiB/s	0.092 ms
Non-ASCII char + ASCII	84.905 KiB	3.84 GiB/s	0.024 ms

Non-chunked, PR

Test	Size	Throughput	Mean Time
Latin lipsum (ASCII)	84.902 KiB	36.19 GiB/s	0.002 ms
Arabic lipsum	79.771 KiB	2.02 GiB/s	0.038 ms
Chinese lipsum	68.203 KiB	4.00 GiB/s	0.016 ms
Arabic + 2 * ASCII	249.577 KiB	3.32 GiB/s	0.075 ms
Non-ASCII char + ASCII	84.905 KiB	4.71 GiB/s	0.022 ms

(should not be affected)

Chunked (1000 byte chunks), 25.5.0

Test	Size	Throughput	Mean Time
Latin lipsum (ASCII)	84.902 KiB	1.31 GiB/s	0.062 ms
Arabic lipsum	79.771 KiB	0.90 GiB/s	0.084 ms
Chinese lipsum	68.203 KiB	0.99 GiB/s	0.066 ms
Arabic + 2 * ASCII	249.577 KiB	1.14 GiB/s	0.209 ms
Non-ASCII char + ASCII	84.905 KiB	1.31 GiB/s	0.062 ms

Chunked (1000 byte chunks), PR

Test	Size	Throughput	Mean Time
Latin lipsum (ASCII)	84.902 KiB	7.45 GiB/s	0.011 ms
Arabic lipsum	79.771 KiB	1.24 GiB/s	0.062 ms
Chinese lipsum	68.203 KiB	1.54 GiB/s	0.042 ms
Arabic + 2 * ASCII	249.577 KiB	2.67 GiB/s	0.091 ms
Non-ASCII char + ASCII	84.905 KiB	7.26 GiB/s	0.011 ms

The benchmark creates about ~70-256 chunks on each test
The improvement is significant

[anonrig]

@ChALkeR can you run benchmark CI that applies to this pull-request?

[ChALkeR]

@anonrig I can't do anything until #61553

Also, we don't have a stream: true TextDecoder benchmark.
The bench results above are from https://github.com/lemire/jstextdecoderbench with this modification:

function decodeChunked(bytes) {
  const chunk = 1000
  const max = bytes.length - chunk
  let i = 0
  for (; i < max; i += chunk) decoder.decode(bytes.subarray(i, i + chunk), { stream: true })
  decoder.decode(bytes.subarray(i))
}

And using that instead of decoder.decode

[anonrig]

Thank you for the kind responses and amazing work.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.77%. Comparing base (e155415) to head (1559954).
⚠️ Report is 38 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff            @@
##             main   #61549    +/-   ##
========================================
  Coverage   89.77%   89.77%            
========================================
  Files         672      673     +1     
  Lines      203755   203875   +120     
  Branches    39167    39190    +23     
========================================
+ Hits       182922   183038   +116     
- Misses      13164    13172     +8     
+ Partials     7669     7665     -4     

Files with missing lines	Coverage Δ
lib/internal/encoding.js	100.00% <100.00%> (ø)
lib/internal/encoding/util.js	100.00% <100.00%> (ø)

... and 33 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/71082/

[ChALkeR]

@anonrig I added a benchmark for streaming Unicode TextDecoder.
Benchmark CI: https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/1787

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/71102/

[ChALkeR]

@gurgunday Once everything is fixed, the plan is to update WPT + bring in my extra tests. They'll fail now due to other bugs though, so it can't be done yet without sorting the tests, and that's likely not worth it if we can just fix stuff first.