test(cli): kill the surviving mutants of tools, telemetry, kernel, runtime and the profile pockets

[blafourcade]

Problem

After #795, four mutation scopes still carry more than a fifth of their mutants alive: tools (818 survived, 219 uncovered of 3379; registry, capabilities, formats), telemetry (844 / 135 of 4413; the per-tool readers), kernel (348 / 58 of 1526) and runtime (194 / 197 of 1289). The profile scopes beside them sit at 80–96, and inside them the copilot buildFilePath handlers (51 survivors), codex mergeCodexConfigToml and the opencode hooks bridge are the remaining pockets.

Scope

• One pass per scope, in the order tools, telemetry, kernel, runtime, then the three profile pockets; each pass its own PR so a floor moves with the tests that earned it.
• Same method as feat(cli): one shared framework source per machine, user scope and session telemetry #795: read the survivors, write the test each one names, red first, restore, argue the rest equivalent with line and reason.
• Floors raised to the measured score minus two.

Acceptance criteria

• tools, telemetry, kernel, runtime each ≥ 85 under run-mutation.mjs <scope> --force
• tools-copilot ≥ 90, tools-codex ≥ 92, tools-opencode ≥ 90
• Red evidence per test in each PR; no source change without a reason

Prior art in this repo

• feat(cli): one shared framework source per machine, user scope and session telemetry #795 and the profiles pass: cli/tests/contexts/tools/domain/profiles/*/build.unit.test.ts.
• Two measurement artefacts to know before counting: stryker records six mutants at plugin-source-tree-reader.ts:39 as survived though a hand-applied mutation turns the test red (perTest attribution across the two vitest projects), and a mutant that makes a capability constructor throw at import is recorded as survived rather than as an error (CapabilityConfigError in the codex profile).

Out of scope

• framework, its own issue.
• Changing what stryker counts; the two artefacts above are reported upstream or lived with, not patched here.