Segfault when calling grossFormula() function

**Summary**
Short summary of problem here

For some molecules, I get a segfault around 30-50% of the time when calling grossFormula() on the molecule. This seems to occur for structures with multiple repeating units, but it does not happen consistently for a given structure.

**Steps to Reproduce**

The following is some code I used to hit this error:

```
from indigo import Indigo

structure = "SCCCOP(OCCOP(OCOCO)(=O)OCC)(=O) |Sg:n:3,4:k:ht,Sg:n:7,8,9:m:ht,Sg:n:8,13,14:n:ht,Sg:n:17,18:k:ht|"
indigo = Indigo()
molecule = indigo.loadStructure(structure)
grossFormula = molecule.grossFormula()
```

**Expected behavior**
A clear and concise description of what you expected to happen.

Retrieve gross formula for molecule without a segfault.

**Actual behavior**
A clear and concise description of what the bug is.

Segmentation fault, looks like it may come from line 211 of `core/indigo-core/molecule/src/molecule_gross_formula.cpp`
```
unit.isotopes.emplace(key, 1);
```

Valgrind output:

```
==6173== Invalid read of size 8
==6173== at 0x5B519A3E: std::_Rb_tree_decrement(std::_Rb_tree_node_base*) (in /home/proj/main/env/lib/python3.11/site-packages/indigo/lib/linux-x86_64/libindigo.so)
==6173== by 0x5AE8167A: std::_Rb_tree_iterator >::operator--() (stl_tree.h:302)
==6173== by 0x5AE80389: std::_Rb_tree, std::_Select1st >, std::less, std::allocator > >::_M_get_insert_unique_pos(int const&) (stl_tree.h:2126)
==6173== by 0x5B130EAF: std::pair >, bool> std::_Rb_tree, std::_Select1st >, std::less, std::allocator > >::_M_emplace_unique(int&, int&&) (stl_tree.h:2434)
==6173== by 0x5B1309CB: std::pair >, bool> std::map, std::allocator > >::emplace(int&, int&&) (stl_map.h:606)
==6173== by 0x5B12F54E: indigo::MoleculeGrossFormula::collect(indigo::BaseMolecule&, bool) (molecule_gross_formula.cpp:212)
==6173== by 0x5AE6E80A: indigoGrossFormula (indigo_calc.cpp:49)
==6173== by 0xBFC8F79: ??? (in /usr/lib/x86_64-linux-gnu/libffi.so.8.1.2)
==6173== by 0xBFC840D: ??? (in /usr/lib/x86_64-linux-gnu/libffi.so.8.1.2)
==6173== by 0xBFC8B0C: ffi_call (in /usr/lib/x86_64-linux-gnu/libffi.so.8.1.2)
==6173== by 0xBF9042A: ??? (in /usr/lib/python3.11/lib-dynload/_ctypes.cpython-311-x86_64-linux-gnu.so)
==6173== by 0xBF8F612: ??? (in /usr/lib/python3.11/lib-dynload/_ctypes.cpython-311-x86_64-linux-gnu.so)
==6173== Address 0x8 is not stack'd, malloc'd or (recently) free'd
==6173==
==6173==
==6173== Process terminating with default action of signal 11 (SIGSEGV)
==6173== Access not within mapped region at address 0x8
==6173== at 0x5B519A3E: std::_Rb_tree_decrement(std::_Rb_tree_node_base*) (in /home/proj/main/env/lib/python3.11/site-packages/indigo/lib/linux-x86_64/libindigo.so)
==6173== by 0x5AE8167A: std::_Rb_tree_iterator >::operator--() (stl_tree.h:302)
==6173== by 0x5AE80389: std::_Rb_tree, std::_Select1st >, std::less, std::allocator > >::_M_get_insert_unique_pos(int const&) (stl_tree.h:2126)
==6173== by 0x5B130EAF: std::pair >, bool> std::_Rb_tree, std::_Select1st >, std::less, std::allocator > >::_M_emplace_unique(int&, int&&) (stl_tree.h:2434)
==6173== by 0x5B1309CB: std::pair >, bool> std::map, std::allocator > >::emplace(int&, int&&) (stl_map.h:606)
==6173== by 0x5B12F54E: indigo::MoleculeGrossFormula::collect(indigo::BaseMolecule&, bool) (molecule_gross_formula.cpp:212)
==6173== by 0x5AE6E80A: indigoGrossFormula (indigo_calc.cpp:49)
==6173== by 0xBFC8F79: ??? (in /usr/lib/x86_64-linux-gnu/libffi.so.8.1.2)
==6173== by 0xBFC840D: ??? (in /usr/lib/x86_64-linux-gnu/libffi.so.8.1.2)
==6173== by 0xBFC8B0C: ffi_call (in /usr/lib/x86_64-linux-gnu/libffi.so.8.1.2)
==6173== by 0xBF9042A: ??? (in /usr/lib/python3.11/lib-dynload/_ctypes.cpython-311-x86_64-linux-gnu.so)
==6173== by 0xBF8F612: ??? (in /usr/lib/python3.11/lib-dynload/_ctypes.cpython-311-x86_64-linux-gnu.so)
==6173== If you believe this happened as a result of a stack
==6173== overflow in your program's main thread (unlikely but
==6173== possible), you can try to increase the size of the
==6173== main thread stack using the --main-stacksize= flag.
==6173== The main thread stack size used in this run was 83886080.
==6173==
==6173== HEAP SUMMARY:
==6173== in use at exit: 95,564,910 bytes in 521,341 blocks
==6173== total heap usage: 2,086,024 allocs, 1,564,683 frees, 628,662,008 bytes allocated
==6173==
==6173== LEAK SUMMARY:
==6173== definitely lost: 192 bytes in 2 blocks
==6173== indirectly lost: 0 bytes in 0 blocks
==6173== possibly lost: 572,047 bytes in 258 blocks
==6173== still reachable: 94,992,671 bytes in 521,081 blocks
==6173== of which reachable via heuristic:
==6173== stdstring : 8,507,488 bytes in 163,222 blocks
==6173== newarray : 226 bytes in 2 blocks
==6173== suppressed: 0 bytes in 0 blocks
==6173== Rerun with --leak-check=full to see details of leaked memory
==6173==
==6173== For lists of detected and suppressed errors, rerun with: -s
==6173== ERROR SUMMARY: 7 errors from 5 contexts (suppressed: 0 from 0)

```

**Environment details:**
- Back-end version: indigo-1.11.0