move fence to the right place

[Maoni0]

fixes #82414

we need to make sure the clearing of the syncblock and the increase of the alloc_allocated not reordered. the comment had the right idea but the implementation was incorrect. the fence should be between these 2 operations, not before the 1st one.

[Maoni0]

/azp run runtime-coreclr gcstress-extra

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[Maoni0]

all gcstress-extra tests succeeded.

https://dev.azure.com/dnceng-public/public/_build/results?buildId=184780&view=results

I'll run it again.

[Maoni0]

/azp run runtime-coreclr gcstress-extra
/azp run runtime-coreclr gcstress-extra

[azure-pipelines]

No pipelines are associated with this pull request.

[Maoni0]

/azp run runtime-coreclr gcstress-extra

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[Maoni0]

/azp run runtime-coreclr gcstress-extra

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[Maoni0]

/azp run runtime-coreclr gcstress-extra

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[ghost]

Tagging subscribers to this area: @dotnet/gc
See info in area-owners.md if you want to be subscribed.

Issue Details

---

fixes #82414

Author:	Maoni0
Assignees:	Maoni0
Labels:	area-GC-coreclr
Milestone:	-

[Maoni0]

/azp run runtime-coreclr gcstress-extra

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jkotas]

Thank you!

[BruceForstall]

It's amazing to see gcstress-extra run clean!

fwiw, the regularly scheduled run today (without your change) failed again with the same asserts on linux/arm32: https://dev.azure.com/dnceng-public/public/_build/results?buildId=185163&view=ms.vss-test-web.build-test-results-tab