Use objc_autoreleaseReturnValue to eliminate the autorelease operation
for return value if the caller is optimized (usually it is).
This required moving autorelease operation from Kotlin -> ObjC ref
conversion to bridge epilogue.
To achieve this, also make ObjCExport Kotlin ref -> ObjC ref dynamic
converters return retained reference (instead of autoreleased one).
To reflect this, rename the corresponding entities in the code.
Unfortunately, llvm removes full debug info from module on any error.
Different version debug info in bitcode is not always compatible, also
it adds this debug info additional requirements on generated debug info.
So this feature is quite unstable and shouldn't be enabled by default,
although it has almost no downsides when worked correctly.
We don't want to deprecate freezing at all, but it is possible
that freezing, new memory model and lazy global initialization
combination might not work in some cases. It might be a problem
when such case appears in 3rd-party library that user can't fix.
To mitigate this problem this commit introduces `freezing` binary
option. It has three variants:
* Full - ol' good behavior.
* Disabled - well, no freezing at all.
* ExplicitOnly - a compromise when user want to freeze something
themselves, but something is messed up during globals initialization.
Calling initRuntime requires some machinery from the
objc module (e.g. adding the toKotlin method to NSObject).
Include this module to the tests to allow full runtime
initialization in the tests
* Do not reset unhandled exception hook
* Add processUnhandledException to perform default unhandled exception
processing
* Add terminateWithUnhandledException to report the unhandled exception
and terminate the program
* Use the default unhandled exception processing in entrypoint, interop
boundaries and in Worker.executeAfter
* Add -Xworker-exception-handling to control exception processing of
Worker.executeAfter. By default its the old behaviour with the old MM,
and new behaviour with the new MM.
Recent versions of LLVM complain when caller and callee have
different vector attributes: https://reviews.llvm.org/D82562.
To mitigate this problem this commit changes how we include
polyHashUnroll* and polyHashTail into Traits: we duplicate them
inside each trait with correct attributes by using
```
#pragma clang attribute push(ATTRIBUTE, apply_to = function)
```
This solution is not pretty for sure, but Clang is not complaining
anymore.
Pavel Kunyavskiy