SEGV in RWMol::commitBatchEdit

[hgarrereyn]

Describe the bug
After certain sequences of adding/removing atoms in a RWMol in batch edit mode, RWMol::commitBatchEdit can segfault. Depending on the call sequence, this either segfaults which crashes the program or performs a heap-buffer-overflow and reads unallocated heap memory (which is only detectible with ASAN).

To Reproduce
This is reproducible in both C++ and Python. Both of the following examples cause a segfault:

#include <GraphMol/RWMol.h>

int main() {
    RDKit::RWMol *mol = new RDKit::RWMol();
    mol->beginBatchEdit();
    mol->addAtom(true);
    mol->commitBatchEdit();
}

from rdkit.Chem import AllChem as Chem

m = Chem.RWMol()
m.BeginBatchEdit()
m.AddAtom(Chem.Atom(1))
m.CommitBatchEdit()

Expected behavior
Based on this test: https://github.com/rdkit/rdkit/blob/master/Code/GraphMol/catch_graphmol.cpp#L1798-L1809 it seems that adding atoms in batch edit mode is allowed behavior.

Configuration (please complete the following information):

• RDKit version: 2021.03.01
• OS: Ubuntu 18.04
• Python version: Python 3.9.2 (although this bug is not dependent on the python version)

Additional context
In an ASAN build of RDKit, the cpp example above crashes with the following stacktrace:

==20==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x7f0a29ce06ca bp 0x7fffca805520 sp 0x7fffca8053a0 T0)
==20==The signal is caused by a READ memory access.
==20==Hint: address points to the zero page.
    #0 0x7f0a29ce06c9 in operator bool /usr/local/include/boost/dynamic_bitset/dynamic_bitset.hpp:104:41
    #1 0x7f0a29ce06c9 in RDKit::RWMol::commitBatchEdit() /src/rdkit/Code/GraphMol/RWMol.cpp:512
    #2 0x519ec9 in main poc.cpp:7:10
    #3 0x7f0a1f558bf6 in __libc_start_main /build/glibc-S9d2JN/glibc-2.27/csu/../csu/libc-start.c:310
    #4 0x41d619 in _start (poc+0x41d619)