[Expression Propagation Memory] Implement propagation of global variables

[mari-mari]

Proposal

Currently, we do not propagate (re)definitions of global variables, as there are cases where it could lead to incorrect decompilation. Consider the following example test_global.zip:

00004010  uint32_t glob = 0x4

propagate_globals:
   0 @ 0000118a  [glob].d = 0xa @ mem#0 -> mem#1
   1 @ 00001194  rax#1 = [glob].d @ mem#1 //<---------it is OK to propagate glob = 0xa here, then rax#1 = 0xa
   2 @ 0000119a  rax_1#2 = rax#1 + 0xa 
...

00004010  uint32_t glob = 0x4

do_not_propagate_globals:
   0 @ 00001159  [glob].d = 2 @ mem#0 -> mem#1
   1 @ 00001163  rax#1 = 0
   2 @ 00001168  mem#2 = change_glob() @ mem#1 //<-------------------- potentially changes glob
   3 @ 0000116d  rax_1#2 = [glob].d @ mem#2 //<-----------------------------cannot propagate glob = 2 here, since function call may change value of glob
....

Our current strategy was not to propagate global variables. However, such propagation would lead to cleaner decompiled code - in cases where it is possible (correct) ;)

Approach

Modify EPM (ExpressionPropagationMemory) to handle also global variables. It can be done in the same fashion as by variables that have pointers to it: check if between definition and use on any path exist instruction that potentially changes the global variable - for global variables, it would be any (non-library?) function call; or, in case pointer on global variable exist, write-uses of this pointer. There could be more corner cases here; they should be handled correspondingly.