fix(trie): InMemoryTrieCursor case where all DB nodes are deleted

[mediocregopher]

This was pointed out by @0x00101010 in a separate PR. Basically if the in-memory overlay completely deleted all nodes in the trie table, then there was no way to distinguish the cases of the db cursor being exhausted or having never been used; the cursor_entry field would be None in both cases.

[Rjected]

This makes sense to me, I wonder if, to make it more explicit, we should have an enum for cursor_entry, which would be:

enum CursorState {
    /// Initial cursor state, when we haven't seeked, open to renames
    Initial
    /// The cursor is currently pointing to an entry.
    Entry {
        /// Path to the node
        path: Nibbles,
        /// The node
        node: BranchNodeCompact,
    },
    /// The DB cursor is exhausted
    Exhausted
}