fix(ModuleLoader): an unparseable unmanaged module version failing the load

[CoffeeFlux]

Problem

When a required module returns a plain version value instead of a DepCtrl record, ModuleLoader.loadModules wraps that value in an unmanaged PackageRecord before it checks the version. If the value doesn't parse, the PackageRecord constructor asserts. The requiring script then fails with:

[DepCtrl] Bad DependencyControl record (Couldn't parse version number: Can't parse version string '##__LOG_VERSION__##'. ...)

The module never reaches the updater, so updating DepCtrl or the requiring script doesn't help. The only fix is deleting the file by hand.

This is what's behind TypesettingTools/Aegisub-Motion#94. The a-mo v1.1.0 release zip shipped untemplated, and a-mo.Log (a plain table with a string version) has been 1.0.0 since 2015, so a broken copy is never replaced. The zip has since been rebuilt, but existing installs stay broken. v0.6.3-alpha had the same assert, so this isn't a regression.

Change

• In loadModules, an unmanaged version that doesn't parse is treated as 0.0.0. It fails any real requirement, so the module goes through the existing outdated-module update path (updater\require).
• In formatVersionErrorTemplate, if that update fails (offline, no feed), the "Installed" value is shown verbatim instead of vnil.
• Added tests for both, plus a changelog entry.

Trade-offs

• A requirement of 0.0.0 (or a packed 0) is still satisfied by a module whose version doesn't parse, so such a module loads silently. It can't be worse than "any version".
• An unmanaged module with a non-semver version (e.g. "1.2", "v1.0.0") that used to hard-fail will now be replaced from the requiring script's feed if one provides it. That's the same thing that already happens to an unmanaged module whose version is merely lower.
• This only covers unmanaged modules. A managed module with a broken version asserts inside its own DependencyControl{} call while it's being required, and still surfaces as a module load error.

Testing

moonc -p passes on both changed files. I couldn't run the suite locally (missing Lua 5.1 rocks), so I'm relying on CI for that.

🤖 Generated with Claude Code

[github-actions]

Test results — macos-latest ✅1116 · ubuntu-latest ✅1116 · windows-latest ✅1113

Summary

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Other ❓	Flaky 🍂	Duration ⏱️
3348	3345	0	3	0	0	1m 8s

_{🎉 No failed tests in this run.}

Github Test Reporter by CTRF 💚

🔄 This comment has been updated

[CoffeeFlux]

I shipped a bad update, and it seems like the recovery path from that is harsher than I thought right now. This should make it possible for maintainers to bump a version and fix these sorts of issues, but needs to end up in a DepCtrl release. Sorry about the hassle.