Trim Cluster A's vendored deployment/upgrade battery

[mkarlesky]

Summary

PR 3 of 5 in the system-test redesign (Stage 2 plan, following the "System Test Duplication Assessment" report's Finding 1 -- the largest single lever, an estimated 17.0% of measured system-test time / ~76 min of full-matrix CI compute per run).

deployment_as_gem_spec.rb is untouched and stays the canonical full CommonSystemTestCases battery (52 examples) -- the one place install-mode-agnostic behavior (preprocessing, Partials, defines, Unity features, crash handling, verbosity, test filtering) runs, under the default (most common real-world) install mode.

Changes

• deployment_as_vendor_spec.rb: 82 → 13 examples. Each of its three ceedling new scaffold variants (--local --docs; --local --docs --gitsupport; --local alone) keeps only its own distinguishing scaffolding-specific checks (does the right vendor/docs/git structure exist) plus one test_project_success smoke proving that variant's vendored install actually builds and runs something for real. Drops version reporting (redundant with the gem spec) and the full preprocessing/Partials/defines/Unity/crash/verbosity/filtering battery -- pure build/runtime logic already proven once, install-mode-agnostic.
• upgrade_as_vendor_spec.rb: 32 → 10 examples. Keeps its only unique payload (can_upgrade_projects/can_upgrade_projects_with_no_test_support_folder -- the only tests that invoke ceedling upgrade at all -- and cannot_upgrade_non_existing_project, the only coverage of the failure path) plus a lean pre/post test_project_success pair bracketing the upgrade, dropping the repeated 8-item generic battery that ran identically before and after upgrade today.

Cluster A total: 166 → 75 examples (55% reduction).

Coverage retention

Verified by comparing test_case symbol sets across the three files: every symbol dropped from vendor/upgrade is either vendor/upgrade-specific (kept in its own file) or still present in the untouched deployment_as_gem_spec.rb. comm -13 between the two symbol sets confirms exactly the 7 expected vendor/upgrade-specific symbols, no unexpected orphans -- nothing silently lost coverage.

Verification

• All 75 examples pass together, host (macOS) and the madsciencelab-plugins Docker image (Linux).
• 4 GDB-related failures seen on host are confirmed pre-existing and environmental: reproduced identically against the unmodified deployment_as_gem_spec.rb on a clean git stash (macOS gdb codesigning issue -- "please check gdb is codesigned - see taskgated(8)"), unrelated to this change. Clean (0 failures) on Linux, where that macOS-specific quirk doesn't apply.
• Full-suite validation (all 16 CI matrix jobs) is CI's job per the Stage 2 plan's process -- not duplicated locally.

This is PR 3 of 5; the plan calls for a go-ahead before starting PR 4 (mixin tier split).

🤖 Generated with Claude Code