test(cli): kill the surviving mutants of the kernel and the runtime

[blafourcade]

🎯 What & why

Fourth and fifth lots of #799. The kernel and runtime mutation scopes held floors of 71 and 67. GitAdapter had no test at all. This PR adds the tests their survivors named. It changes no source.

Stacked on #818. The runtime lot found a real defect: telemetry off dropped the executable bit of a person's own commit hook. Its fix ships on its own in #818, and this branch sits on it. Review #818 first. Once it merges, this PR's diff against next is tests only.

🛠️ How it works

Scope	Before	After	Floor
kernel	75.2 (340 survived, 38 uncovered)	97.1 (45 survived, 0 uncovered)	71 → 95
runtime	69.8 (199 survived, 193 uncovered)	92.2 (88 survived, 13 uncovered)	67 → 90

• Kernel: every error class's exact name and message, plus the markdown, semver, source, file, merge, paths, and JSON/JSONC reading helpers.
• Runtime: the first GitAdapter test, on a real temporary repository. The auth adapters, self-updater, file adapter, prompter, HTTP client, update check and release resolver are asserted on exact commands, messages and headers, and so is the composition root.
• Red first: each test was red against a mutant applied by hand, 251 in kernel and 137 in runtime.
• What survives is written down: each remaining survivor is noted with its reason. Most are equivalent mutants, the rest composition-root literals with no observable seam.
• Two POSIX-only checks skip where the platform cannot show them: a mode bit on Windows, and a read-only directory on Windows or as root.

🧪 How to verify

git diff --stat origin/fix/telemetry-off-keeps-hook-executable... -- cli/src   # empty
cd cli && pnpm vitest run --project unit --project integration tests/kernel tests/runtime
node scripts/run-mutation.mjs kernel && node scripts/run-mutation.mjs runtime

Gate	Result
kernel and runtime suites, after the rebase	68 files, 808 tests passed
typecheck, test:arch, knip	clean
full suite, before the rebase	5381 tests passed

⚠️ Heads-up

• CI runs the kernel and runtime mutation gates on this PR.
• The three symlink tests in file-adapter-edges are not skipped on Windows. The Windows job decides whether they need to be.
• Two infinite-loop mutants in jsonc.ts are killed by stryker's timeout, not by an assertion.

🔗 Linked issue

Refs #799. Only the profile pockets remain after this.

✅ I certify

• I have read the contributing guide and followed the PR template.
• Tests were written red first and every gate listed above passed locally.

🤖 Generated with Claude Code

https://claude.ai/code/session_011x4ms5qcGuZgYhCxfdHMUb

[blafourcade]

Extra commit 2c33f9e: stacked with #815, the two PRs put tests/ at 2989 comment lines against the 2987 baseline. Each one passed alone. Naming the repeated mode-bit skip once brings the stack to 2986, with no baseline raise. Found on a local branch stacking all the open PRs.