[native-mt, 1.5.0-RC] Fix withTimeout throwing InvalidMutabilityException.

[andersio]

Resolves #2688.

TimeoutCoroutine subclasses ScopeCoroutine and adds a backing field time.

Since ScopeCoroutine may potentially freeze itself during its init block (by adding itself to a frozen job as a child), the init steps of TimeoutCoroutine as a subclass may consequentially fail, due to the object now potentially failing MutationCheck().

• Introduced a useInitNativeKludge constructor parameter in ScopeCoroutine to control whether the init kludge for native is enabled.

• Fixed InvalidMutabilityException raised by withTimeout {} when the scope is frozen (1.5.0-RC-native-mt) #2688 by changing TimeoutCoroutine to disable the superclass' kludge, in favour of having the kludge as its own init block.

• Added a failing test case.

  • The test case suffers from issue (2) in InvalidMutabilityException raised by withTimeout {} when the scope is frozen (1.5.0-RC-native-mt) #2688 though, that the test runner should have been reporting the InvalidMutabilityException. But instead, runBlocking() eats the exception and continues to spin indefinitely. I didn't dig into why this happens. It looks like a consequence of (unexpectedly?) failing during initialisation, which failed child does not remove itself from the parent.

[ArcticLampyrid]

if (!isReuseSupportedInPlatform()) initParentForNativeUndispatchedCoroutine()

Personally, I prefer to make this as a initParentJob method in the superclass for maintainability

[qwwdfsad]

Great contribution, thanks for finding the root cause and the fix!

Hopefully we'll have a new memory-model soon and won't be needing to maintain such kludges