chore: improve performance of scheduling effects (but without creating additional branches)

[Rich-Harris]

Another take on #13282 — rather than creating an additional branch effect for every effect_root, this just flags roots as being queued and avoids queuing them a second time. On flush, the flag is removed.

[changeset-bot]

🦋 Changeset detected

Latest commit: e200487

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
svelte	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[trueadm]

Awesome :)

[trueadm]

Ah, with this PR I get this error when running the reactivity benchmark locally running all the benchmarks together.

RangeError: Maximum call stack size exceeded
    at check_dirtiness ([stdin]:823:25)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)
    at check_dirtiness ([stdin]:841:13)

If I remove the EFFECT_QUEUED logic or move the the bit where we change the flag off to after the flushing then the benchmark works (but our tests fail). I'll need some time to dig into this.

Update: I was only running a subset of the benchmarks. Running them all together and it seems to work fine?

[trueadm]

Removing my approval for now

[trueadm]

So I think there was an issue with how I was benchmarking, it seems to be fine now.