v8: use AliasedBuffers for passing heap statistics around

[joyeecheung]

Instead of holding shared pointers to ArrayBuffers, simplify
the code by using AliasedBuffers directly which allows the
binding to own the buffers.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/30840/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/30845/

[mscdex]

This seems to cause a performance regression for getHeapSpaceStatistics() specifically:

                                                          confidence improvement accuracy (*)   (**)   (***)
v8/get-stats.js n=1000000 method='getHeapSpaceStatistics'        ***    -12.75 %       ±2.36% ±3.14%  ±4.09%
v8/get-stats.js n=1000000 method='getHeapStatistics'                     -0.05 %       ±5.79% ±7.72% ±10.07%

[jasnell]

LGTM but I've been thinking strongly about moving things away from AliasedBuffer to the new AliasedStruct to give us a bit more natural way of interacting with the data on the native side.

For instance, instead of having to use buffer[idx] = val on the native side, we'd have something more like buffer.foo = val

[joyeecheung]

@mscdex yes, the performance regression is caused by more calls into the C++ in getHeapSpaceStatistics() - though this is unlikely to be on any hot code path, so I think in this case the simplicity of the code is more important.

@jasnell Thanks for the heads up, it seems #32778 landed fairly recently. It seems to make more sense than the current homogenous array-based approach to me, though I guess we can make the transition incrementally :)

[mscdex]

Perhaps we should remove benchmark/v8/get-stats.js then?

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/31043/

[addaleax]

Landed in aa9708e