Perform HASHTAB_ITER bookkeeping before user-code

The `HASHTAB_ITER` logic keeps track of how many entries in the hash
table are left to visit, decrementing this on each iteration of the
loop. This was previously decremented at the end of the loop body:

```c
size_t hi##todo_ = hi##ht_->ht_used;
for (hashitem_T *hi = hi##ht_->ht_array; hi##todo_; hi++) {
  if (!HASHITEM_EMPTY(hi)) {
    {
      <user code, macro-inserted>
    }
    hi##todo_--; // <--- important decrement here
  }
}
```

This meant that if the body of the loop (substituted in via macro
expansion) contained a `continue` statement, we'd skip decrementing our
counter, meaning we'd iterate too many times over the hash table,
usually leading to an out of bounds read beyond the hash table's
memory, or uninitialised/null pointers from unused hash table slots.

Decrementing `hi##todo` before the arbitrary loop body protects us from
this, and has no adverse side-effects since only the macro code can (or
should) use this variable.

Before this commit, no code within `HASHTAB_ITER()` contained a
`continue`, meaning this bug was left dormant and the fix has a very
minimal chance of introducing any bugs.

src/nvim/hashtab.h

@@ -81,10 +81,10 @@ typedef struct hashtable_S {
       size_t hi##todo_ = hi##ht_->ht_used; \
       for (hashitem_T *hi = hi##ht_->ht_array; hi##todo_; hi++) { \
         if (!HASHITEM_EMPTY(hi)) { \
+          hi##todo_--; \
           { \
             code \
           } \
-          hi##todo_--; \
         } \
       } \
     } while (0)