[PATCH] [analyzer][Bugfix] RegionStore: use pointee type to create one-element regions

[Aleksei Sidorin]

Here is a sample test case:

```
char passAndModifyPtr(int *p) {
  if (*p > 10) {
    *p = 4;
    return 0;
  } else {
    if (*p < 5) {
      *p = 7;
      return 1;
    }
  }
  return 2;
}
```

And this code should be inserted in checkDeadSymbols() callback of some checker:
  for (SymbolReaper::dead_iterator I = SymReaper.dead_begin(),
      E = SymReaper.dead_end(); I != E; ++I) {
    SymbolRef Sym = *I;
    Sym->dump();
    llvm::errs() << "\n";
    // One of the symbols reaped is a value associated with *p
    if (const SymbolRegionValue *V = dyn_cast<SymbolRegionValue>(Sym)) {
      const MemRegion *MR = V->getRegion();
      // one-element ElementRegion is created automatically
      if (const ElementRegion *ER = dyn_cast<ElementRegion>(MR)) {
        const MemRegion *BaseReg = ER->getBaseRegion();
        // BaseReg is a SymRegion{reg_$0<p>}
        // Its value should be nonloc::SymbolVal(Sym)
        // i.e. reg_$1<element{SymRegion{reg_$0<p>},0 S32b,int}>
        BaseReg->dump();
        llvm::errs() << "\n";
        State->getSVal(BaseReg).dump();// but wrong SVal is created here
      }
    }
  }
SVal related to `*p` is `nonloc::SymbolVal(reg_$1<element{SymRegion{reg_$0<p>},0 S32b,int}>)`. The same value should be retrieved by `getSVal(SymRegion{reg_$0<p>})`. But the result of this action is `&SymRegion{reg_$2<element{SymRegion{reg_$0<p>},0 S32b,int *}>}` which seems to be incorrect.

http://reviews.llvm.org/D4974