fix: require pooled Android benchmark emulators

[janicduplessis]

Description

Android benchmarks still create a fresh AVD for Stim, so they cannot measure emulator reuse. That policy dates to the original Android benchmark, before Android pooling.

Solution

Prepare one parked Android emulator and require Stim to adopt that exact device, matching the existing iOS methodology. Reject missing, running, or incompatible pool state before dispatch and verify the emulator returns to the pool after cleanup. Control still creates a fresh device. Old Android golden state must be prepared again; published results are not changed by this PR.

Test plan

• Regression checks reject disabled/missing pools, changed AVD configuration, a running or deleted device, and a different Stim AVD with the same model/image.
• The candidate passed a real Android pool lifecycle smoke test on API 36 arm64, including app-data reset and retained-APK installation skipping.
• Campaign preparation and dispatch remain untested until publication.

Fixes #573.