Use a weak event handler pattern for GlobalOptionService

[sharwell]

Force the use of a weak event handler pattern for GlobalOptionService. In the event a PreviewWorkspace fails to be released, it will no longer be GC rooted through this event.

Extracted from #68054

[CyrusNajmabadi]

Done with pass.

[tmat]

Seems odd to have to change all global option change handlers because we have an issue with one of them.

[sharwell]

Seems odd to have to change all global option change handlers because we have an issue with one of them.

We never want GlobalOptionService to be the GC root of an item in the graph. A secondary benefit of this change is GC Root commands in the future will automatically de-prioritize the weak reference from GlobalOptionService and better reveal the true owners of objects in a heap.

[tmat]

new();

How much overhead is there for each CWT instance? Would there be benefit in making this static?

[sharwell]

➡️ It could result in some unexpected semantics if the table overlapped (e.g. removing the event handler from one event might remove the keep-alive status for a different event).

Note that in the current code, only one of these will exist (GlobalOptionService is a singleton).