DATAS BGC thread synchronization fix

[Maoni0]

this problem with BGC thread synchronization is similar to the SVR GC one I fixed but has its own complications due to the fact BGC threads are only created on demands. this was work I wanted to do in .NET9 but didn't get time to.

it can cause the following symptoms -

• deadlock because a BGC thread is erroneously terminated so then the next BGC join cannot finish
• AV because a BGC thread could be seeing settings.concurrent as true and actually running blocking GC code!

I moved setting the idle event code to be inside a GC instead of outside because we don't necessarily trigger a BGC in-between HC changes. and if we don't, we'd need to compensate for the idle count that we deducted and this is difficult to track. moving this to the place where we already decided to do a BGC makes it much simpler. I am setting all the required idle events sequentially instead of per heap (there isn't a convenient way to set it per heap without introducing yet more sync cost) which does incur some perf cost - however this cost is small and we only need to pay for it when the following conditions are true - 1) we actually changed HC; 2) we did do a BGC that observed this HC change; 3) we need to wake up threads that participated in the last BGC.

also added some stress code to help with reproing the problem and stressing the new code.

[dotnet-policy-service]

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

[mangod9]

lgtm. Thanks for fixing.

[Maoni0]

last commit I pushed will be cleaned up - I left some code there to help with stress.