Profile: make heap snapshots viewable in vscode viewer

[IanButterworth]

• Fixes Heap snapshots aren't viewable in vscode viewer #52680
• Tidies up streamed files when not explicitly using streamed mode

Tested against https://github.com/microsoft/vscode-v8-heap-tools/tree/main/v8-heap-parser by copying a julia heapsnapshot into the test dir as basic.heapsnapshot and running cargo test

Update: This is working, except for the graph view, which appears to be an upstream issue microsoft/vscode-js-profile-visualizer#175

---

Original issues

With these changes, it still fails, but because of the middle line here

"codelocs",
"�\u0003",
"<generic memory - malloc>",

with

failures:

---- decoder::tests::test_basic_heapsnapshot stdout ----
thread 'decoder::tests::test_basic_heapsnapshot' panicked at v8-heap-parser/src/decoder.rs:574:33:
expect no errors: Error("invalid unicode code point", line: 6020770, column: 9)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

---- graph::tests::test_retained_sizes stdout ----
thread 'graph::tests::test_retained_sizes' panicked at v8-heap-parser/src/graph.rs:719:45:
expect no errors: Error("invalid unicode code point", line: 6020770, column: 9)


failures:
    decoder::tests::test_basic_heapsnapshot
    graph::tests::test_retained_sizes

If I manually empty that string the only test that fails is a content comparison against a fixed reference.
Does the string sanitizer need fixing or is this some other issue?

However.. the string-fixed file loads in neither chrome devtools, nor vscode's viewer.

I'm opening this in case you see the issue, @vilterp @NHDaly

[vtjnash]

JavaScript JSON strings differ slightly from Unicode encoding in that high code points are written as malformed UTF-16 surrogate pairs (when encoded as \x or \u escape sequences) instead of being encoded as their Unicode value (which is only supported in the very latest new version of ecmascript). Julia also may include arbitrary binary data in a string, while JSON only permits valid UTF-8 sequences (even as escape sequences)

[IanButterworth]

I see. I just coincidentally pushed a change to only print unicode assigned chars. Does that sound reasonable?

With this they open in vscode now, although the graph view either takes longer than I've tried to build, or hangs. It's not clear

[IanButterworth]

Note that vscode is on an older version of https://github.com/microsoft/vscode-v8-heap-tools/tree/1e1cbd86d9376995e459c31fbaacbd0f7a03adb9
Testing on the right commit helped..

[IanButterworth]

It turns out that opening a heapsnapshot in the graph viewer via the dialog isn't valid and will be fixed (disabled) in microsoft/vscode-js-profile-visualizer#179

Instead, clicking on the graph icon where presented in the table view beside an object opens the graph successfully.

@vilterp @NHDaly I'll merge this tomorrow if I don't hear otherwise.

[vilterp]

Oh, so it's just a 'retainer graph'? i.e. starting from a given object, go up to the roots? that's cool

[vilterp]

lgtm!

[IanButterworth]

Just remembered I needed to update stdlib/Manifest.toml because Profile took on Unicode as a dep.

[vtjnash]

I had thought jl_static_show_string from #52799 would have already fixed this, but I didn't realize that this code used a different string encoder that is somewhat more specific to JSON output. That change might still affect the places where jl_static_show is used here (e.g. for printing some Type representations) but not commonly so