fix(node): seperate worker module cache

[littledivy]

Construct a new module graph container for workers instead of sharing it with the main worker.

Fixes #17248
Fixes #23461

[dsherret]

How's the memory usage? Does it reuse the Arc<str> when it can? (ex. for remote modules)

Also, what happens with --reload?

[dsherret]

@littledivy do you think we could reduce memory usage by sharing the strings in memory when they're the same?

[littledivy]

yeah, maybe we could store specifier -> hash mapping in a shared container and compare that in the file fetcher. Doing this check without comparing specifiers first may be expensive but more effective in some cases.

[satyarohith]

nit: it would be nice to convert this test into a step under worker_threads folder instead of creating worker_threads_cache folder.

[satyarohith]

Hmm. I thought it would be clean to convert several tests of a module into steps but it will break reporting.

[satyarohith]

After #23667, you can now declare serveral tests in __test__.jsonc itself.

[bartlomieju]

Drive-by: can we add tests for both Web and Node workers that verify this behavior?

[dsherret]

Looks like some wpt tests are failing now.

[littledivy]

Looks like some wpt tests are failing now.

It's a new test thats passing now.

 "/html/semantics/scripting-1/the-script-element/module/dynamic-import/blob-url-workers.window.html - A revoked blob URL will not resolve in a worker even if it's in the window's module graph"

[bartlomieju]

Other issues that might potentially get fixed by this one:

• web workers - import() is not cached? #21035
• Memory leak(?) when working with Workers #18414
• Worker doesn't refresh after file changes #20104

[dsherret]

See #23634 (comment)

[dsherret]

LGTM. Seems like there's some errors after merge though.

(We will circle back to add a string cache later)