to_mmcif: quote empty pdbx_strand_id so _struct_ref_seq stays readable

[martinemnoble1]

Problem

For a structure with a blank chain name (e.g. a PDB file with no chain ID) and a DBREF, make_mmcif_document() produces an mmCIF file that gemmi itself cannot read back:

import gemmi
st = gemmi.read_structure('1pfe.cif.gz')
for ch in st[0]:
    if ch.name == 'B':
        ch.name = ''          # blank chain name + existing DBREF
doc = st.make_mmcif_document()
gemmi.cif.read_string(doc.as_string())
# ValueError: Wrong number of values in loop _struct_ref_seq.*

The empty _struct_ref_seq.pdbx_strand_id is written as a zero-width token, so the data row ends up with one value fewer than the loop has tags. The loop is ragged, and the whole file becomes unreadable — not just by gemmi's strict reader, but by other consumers too (this surfaced as an MMDB/clipper Message_fatal abort in a downstream pipeline reading gemmi-written mmCIF).

It reproduces on current master and on the 0.7.5 release. Real-world trigger: legacy single-chain PDB entries with a blank chain ID that carry DBREF records (e.g. PDB 2ACY).

Cause

In the _struct_ref_seq row builder (src/to_mmcif.cpp), every field is passed through a sanitising helper (string_or_qmark, string_or_dot, .str(), qchain) — except pdbx_strand_id, which was added raw:

seq_loop.add_row({std::to_string(++counter2),
                  std::to_string(counter),
                  strand_id->second,  // pdbx_strand_id  <-- raw; empty -> zero-width token
                  id,
                  ...

When strand_id->second is empty, nothing is emitted for that column.

Fix

Wrap it in qchain() — exactly what the adjacent _struct_ref row already does for qchain(ent.name), and what qchain's own comment describes (chain names that may be empty, written as ''). The empty value now serialises as '' (a valid quoted empty string), the loop is well-formed, and it round-trips.

qchain(strand_id->second),  // pdbx_strand_id

One-line change. Added a regression test (tests/test_mol.py::test_empty_strand_id_struct_ref_seq) that blanks a chain name on 1pfe and asserts the written _struct_ref_seq re-reads cleanly; it fails before the change (Wrong number of values in loop _struct_ref_seq) and passes after. Full test_mol.py/test_cif.py pass with no regressions.