[aozalevsky]

Relates #1080.

Two small changes that make pickling use less memory, and make it faster for density maps, without changing the pickle format:

1. _get_as_binary() serialized into an ostringstream, then copied the data into a std::string and again into a bytes object; _set_from_binary() copied the bytes into a std::string and again into an istringstream. The data are now copied once, from the stream buffer into the bytes object, and read directly from the bytes object (two small helper classes in swig_helpers_base.h).
2. DensityMap::serialize() wrote and read the voxel data one value at a time; it now does so with a single cereal::binary_data call.

Pickling a 256³ IMP.em.DensityMap (134 MB) and an IMP.Model with 10⁶ XYZR particles (41 MB); Release build, Python 3.11, median of 3; temporary memory as a multiple of the pickle size:

	develop	this PR
DensityMap dumps	0.36 s, 2x	0.14 s, 1x
DensityMap loads	0.22 s, 3x	0.11 s, 1x
Model dumps	0.14 s, 2x	0.14 s, 1x
Model loads	0.89 s, 3x	0.89 s, 1x

Pickling 100,000 Vector3Ds is 5-8% faster. Pickles are byte-for-byte the same as before (checked for Model, a restraint, VersionInfo, Vector3D and DensityMap), so old pickles load with this PR and vice versa.

Protocol 5 out-of-band buffers (PEP 574), also suggested in #1080, are left for a follow-up. A prototype that wrapped every object's state in a PickleBuffer from a Python __reduce_ex__ made pickling 100,000 Vector3Ds with protocol 5 slower (0.16 s to 0.26 s), so out-of-band buffers should probably be used only for large objects; and reading them without a copy needs Py_buffer, which is only in the limited API from Python 3.11 (the Windows installer is built against the 3.10 limited API).

Existing tests cover the changed code (e.g. the pickle tests in modules/em/test/test_density_map.py and modules/kernel/test/test_version_info.py). The CI jobs pass locally (Ubuntu 24.04 Release and Debug, Ubuntu 26.04 Fast, clang standing in for macOS), as does a Fast build with IMP_PY_LIMITED_API=3.10, as used for the Windows installer.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.72%. Comparing base (318a8ec) to head (cf988ac).

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1127      +/-   ##
===========================================
- Coverage    67.72%   67.72%   -0.01%     
===========================================
  Files         1811     1811              
  Lines       154809   154809              
  Branches     18519    18520       +1     
===========================================
- Hits        104852   104850       -2     
- Misses       45245    45249       +4     
+ Partials      4712     4710       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.