Fix executor context propagation race with stateless lambdas

[laurit]

Resolves #14805
When the same instance is passed to executor multiple times the context propagation virtual field gets overwritten and reset which makes context propagation unreliable. This issue is now easier to trigger after #14546 because previously some executor methods would wrap the given instance into another runnable that would also carry context. Even when the original shared instance had the context reset the wrapper still carried it. #14546 disabled propagating the context into the wrapper. We have previously already fixed this issue for execute(Runnable). This PR applies the same approach to other methods like submit(Runnable) and submit(Callable). Note that this fix only applies to tasks that are stateless lambdas. If a regular singleton is passed to executor it will have the same issue. To fix that we'd need to wrap all tasks similarly to what was proposed in #9325 I don't remember why that PR was never merged.

[trask]

To fix that we'd need to wrap all tasks similarly to what was proposed in #9325 I don't remember why that PR was never merged.

Yeah, me either, seems worth revisiting