Backport fixes and features from the ipfedor fork

[gsokoll]

The fork https://github.com/ipfedor/codescribe (@ipfedor) is 13 commits ahead of main and contains several genuine fixes and features developed against non-IFM hardware (Xinje XS Studio OEM CODESYS) and large CFC/LD projects. After review, we plan to adopt the following by reimplementation, with credit to the fork in each PR:

Bug fixes (one PR)

• Visualisation/POU name collision: a visualisation named like a POU exports to the same Name.xml and one overwrites the other. Fix: export visualisations as Name.vis.xml (fork commit 2e93ff4).
• Mutation during iteration in remove_tracked_communication_devices.
• Missing GUID mappings: persistent GVLs (261bd6e6) and methods without a return type (f89f7675) are currently skipped silently on export. Plus named entries for INTERFACE, TASK, IMAGEPOOL, TEXTLIST, GLOBAL_TEXTLIST, PROPERTY_METHOD.
• Silent skip of objects with unmapped GUIDs during export (fork hit this with an LD nested inside a CFC, commit 7eaec7d): add a native-export fallback with a warning.
• project_template.py uses scriptengine.projects.primary instead of its project parameter.
• DUT import whitespace normalization for idempotent round-trips.

Staging-folder export (one PR)
Export to a staging folder and swap on success instead of rmtree up front, so a locked folder (Explorer, git, antivirus) or a mid-export failure no longer destroys the on-disk copy (fork commits bc6ec89, 1141d7d).

Device-tree siblings export/import (one PR)
Track Ethernet/Modbus/fieldbus devices that sit next to Plc Logic in the device tree (standard CODESYS layout, vs the IFM-style Communication node we already handle), exported to Device/devices/<name>.xml (fork commit 6e08a99).

Library export command (one PR)
Export Lib To Files for library projects, where objects hang off the project root rather than under a Device (fork commit a90c84a).

Not adopting for now: the external XML to ST converter integration, mojibake path repair, OEM install detection, and the gc/retry OutOfMemoryException workarounds. Reasons in the PRs.

@ipfedor thank you for the work in your fork. We are reimplementing rather than merging directly because our CI requires ASCII-only source and the on-disk formats have diverged, but review and testing of the PRs referencing this issue would be very welcome.

[gsokoll]

All four adopted items are now merged to main and this issue is complete.

PR trail

• Bug fixes: Backport bug fixes from the ipfedor fork #25
• Staging-folder export: Stage exports and swap on success instead of rmtree up front #26, re-landed on main via Re-land staging-folder export on main (PR 26 merged into its stacked base) #29 after a stacked-merge mishap
• Device-tree sibling devices: Export and import device-tree sibling devices #27, continued and merged as Export and import device-tree sibling devices #30
• Export Lib To Files: Add Export Lib To Files for library projects #28, continued and merged as Add Export Lib To Files for library projects #31

Everything was validated live on CODESYS V3.5 SP11, including export/import round-trips and a hardware download to an IFM CR711s. Full evidence is in the PR comments.

One finding worth offering back, @ipfedor: while validating the device-tree feature on IFM hardware we hit two flaws that your fork shares. Children of package-fixed devices (Local_IO/Inputs, HMI/User_LEDs on IFM controllers) cannot be removed, so the import aborts, and import_native pastes the archived object as a child of the receiving node rather than replacing it in place, so importing a device xml onto the device itself produces a rejected duplicate paste (HMI_1). Our redesigned import in #30 removes the matching device and imports the xml into the PLC device, recreating it, and skips package-fixed devices whose configuration the project template carries. That change may be worth porting back to your fork if you ever run it against hardware with fixed device trees.

For the record, the items we deliberately did not adopt remain as listed above: the external XML to ST converter integration, the mojibake path repair, the OEM install detection, and the gc/retry OutOfMemoryException workarounds.

Thank you @ipfedor for the work in your fork. Review and testing of what landed here is welcome any time.