Fix handles being split by asset_server_managed.

[andriyDev]

Objective

• Fixes Dropping a handle from Assets::get_strong_handle has different behaviour to one from AssetServer::load. #20651.
• Alternative to unifying Asset<A> / AssetServer handles (#20651, #20776) #20874.
• Also a step towards Do not unload subassets or the root asset unless all subassets and root asset handles have been dropped. #21878.

Solution

• When allocating a new handle (whether for an existing asset or a new asset), send a "new" message to the same channel as "drop" events.
  • This essentially means we're "leaning in" to the fact that we can have multiple live handles to the same asset.
• Assets::track_assets now increments when a "new" message is received, and decrements when a "drop" message is received. The asset entry is dropped if the counter is at 0 after processing all messages.
• Remove the asset_server_managed flag, since we can just lookup the index to see if it's present.
• Remove the handle_drops_to_skip counter since now we count all new handles, not just those in get_strong_handle.

Testing

• Unbroke our reproduction test.
• Added a new test to verify that "reacquiring" an asset doesn't drop the asset.

[kristoff3r]

LGTM, the logic in track_assets felt a bit complicated but I couldn't find a way to simplify it.

[kristoff3r]

I don't see why this import needs to be inside this function.

[andriyDev]

This file defines a type Entry, so the hashmap Entry collides with it. Scoping it here sidesteps that issue.