mapConcurrently and mapConcurrently_ behave differently than corresponding functions in the async library

[torgeirsh]

Consider this simple test:

import qualified Control.Concurrent.Async as Async
import qualified Control.Exception as Exception
import qualified UnliftIO.Async as UnliftIO

main :: IO ()
main = UnliftIO.mapConcurrently_ (forever . print) ("01234" <> undefined :: [Char]) `Exception.finally` putStrLn "Finished"

The expectation is that the threads spawned by mapConcurrently_ are terminated before "Finished" is printed, but this is the output:

$ ./utest
Finished!
'3'
'4'
'0'
'1'
'2'
utest: Prelude.undefined
CallStack (from HasCallStack):
  undefined, called at utest/Main.hs:9:64 in main:Main
'3'
'4'

Replacing "UnliftIO.mapConcurrently_" with "Async.mapConcurrently_", I get the expected output:

$ ./result/bin/log-performance-test 
'1'
'0'
Finished!
utest: Prelude.undefined
CallStack (from HasCallStack):
  undefined, called at utest/Main.hs:9:61 in main:Main

The mapConcurrently (without _) function has similar differences. The test was run on Ubuntu Linux 23.10 with GHC 9.4.8, async 2.2.4, and unliftio 0.2.25.0.