fix(server): registry test stubs no longer outlive the test run

[yordis]

Problem

The fake claude CLI spawned by ProviderInstanceRegistryLive.test.ts keeps an unconditional setInterval(() => {}, 1_000) alive. The probe aborts the SDK without awaiting the child, so the stub is orphaned (reparented to PID 1) and keeps running until the machine reboots. Every run of this test strands more idle node processes.

Change

The stub clears its interval and exits once stdin closes, the same pattern testing/ClaudeCapabilitiesProbe.fixture.mjs already uses.

Scope and approval

Very small, focused fix for an obvious bug: one test fixture in one file, no product behavior touched. It only lets a stub process exit once its parent is gone.

Verification

• Before (unmodified main): cd apps/server && vp test run src/provider/Layers/ProviderInstanceRegistryLive.test.ts passed 8/8, but ps -axo pid,ppid,command | grep t3-provider-path-test afterwards listed 5 orphaned node .../claude processes with PPID 1. The process list was empty before the run.
• After: same command, 8/8 pass, and the same ps check is empty.
• Lint scoped to the file is clean.
• Not checked: other providers' test fixtures for the same pattern, since this PR covers only the leak reproduced here.

[macroscopeapp]

Approvability

Verdict: Approved at cf20b96

Macroscope's review found this PR approvable — This is a one-file, test-only fix that cleans up a spawned fake CLI process after the test ends. All changes are confined to ignored paths and have no product, default, deployment, security, billing, or static-analysis impact.

Notes:

• No code objects were reviewed. Approvability was decided on eligibility alone.

^{You can add or adjust custom eligibility rules. Learn more.}