Cut tests that can't fail and make weak ones catch regressions

An audit of all 630 tests found 25 that guard nothing: they assert a
constant equals itself, print a benchmark, or repeat a stronger test on
the same code path. They are removed.

Eighteen more guarded real behavior but could not fail. The largest:
the .ica round-trip tests imported and exported through copies of the
production code kept inside the test file, so a broken importer still
passed. They now go through loadFromFilePath and zipStrategy against a
throwaway Hive library, every fixture round-trips, and corrupt files
must fail without saving. Skipping migration in the real importer now
fails the fixture test; before, it passed.

The rest: assertions too loose to fail, expected values computed by the
code under test, setup that never reached the branch named in the
title, and a Windows-only provider path that could not tell its two
entry points apart.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
98bf450eedDara Adedeji committed on 10/1/2026, 10:10:02 PMparentd8f2a2d
32 files changedLine totals unavailable