When using `event_manager` to initialize the crawler, the global `event_manager` doesn't work.

[Mantisus]

If event_manager is passed as an initialization parameter to BasicCrawler, the global event_manager in service_locator will never be started.

https://github.com/apify/crawlee-python/blob/master/src/crawlee/crawlers/_basic/_basic_crawler.py#L770-L778

As a result, classes that use service_locator.get_event_manager() do not behave as expected, because they do not receive the relevant events.

Classes affected by this issue: Snapshotter and RecoverableState.

Related: #1805

Also, if two crawlers share the global event_manager and one finishes before the other, the event_manager will be shut down for the second crawler as well.

[Pijukatel]

Just for the discussion. I think we can split this into two problems.

Reliance on global

(We should probably first think through whether there is a scenario for multiple active event_managers that is valid and makes sense.)

Some parts of our code rely on the global event_manager no matter what. Even when they work in the context of another event_manager

Ideas?

• All such event_manager consumers should have explicit event_manager argument, probably on init and thus remove the reliance on the global.

Sharing the same event_manager

• Using it as a reentrant context manager despite the context manager not being designed to handle it this way
• Letting individual consumers close it even though multiple consumers can rely on it

Ideas?

• Having isolated contexts instead of shared ones and avoiding reentrant scenarios:

event_manager = EventManager(...)
async with event_manager.create_context():
   ...

• Handle the re-entrant scenarios by counting consumers