Run LIT tests without optimization

[mhasel]

When compiling with optimization, LLVM will sometimes optimize out faulty IR we generated. The same IR can lead to segfaults when compiling without optimizations - this leaves quite a large gap in our automated test-coverage and will cause bugs to slip through.

Here's an example (which is already on master, running - and passing - with every test-run) I stumbled upon while working on initializing temporary variables:

rusty/tests/lit/single/init/function_locals.st

Lines 1 to 27 in cbe548d

	// RUN: (%COMPILE %s && %RUN) | %CHECK %s
	VAR_GLOBAL
	ps: STRING := 'Hello world!';
	ps2: STRING;
	END_VAR
	FUNCTION_BLOCK foo
	VAR
	s AT ps: STRING;
	END_VAR
	VAR_TEMP
	s2 AT ps2: STRING;
	END_VAR
	s2 := 'Hello from me, too';
	// CHECK: Hello world!
	printf('%s$N', REF(s));
	// CHECK: Hello from me, too
	printf('%s$N', REF(s2));
	END_FUNCTION_BLOCK
	FUNCTION main: DINT
	VAR
	fb: foo;
	END_VAR
	fb();
	END_FUNCTION

This lit test will pass the CI without problems, however, if the same code is compiled with -Onone it leads to a segfault when executed.

We should update our runner so that all LIT tests are ran with default optimization as well as with -Onone.