test: add offerings startup perf tester app

[vegaro]

New test-apps:perftester app to measure how long configure() + getOfferings() takes at startup, baseline vs workflows. It's the realistic counterpart to the instrumentation harness: a real app, a button press, timestamps around the two calls, one JSON line appended per press to results.jsonl on device.

Why an app and not just the instrumentation test: the harness runs in-process with mocked billing and a simulated lifecycle, so it can't show real process cold-start ordering. This can. The had_* cache flags in each row say whether a sample was cold or warm, so run order doesn't have to be tracked separately.

How it works:

• Toggle picks the arm: baseline or DangerousSettings.forWorkflows().
• "Configure + getOfferings" runs one measured cycle: close() if configured (so configure() doesn't short-circuit on an equal config), snapshot the cache flags, configure(), getOfferings(), then on the workflows arm ui_config + first workflow body readiness.
• "Clear SDK caches" wipes the SDK prefs/dirs for a cold sample without reinstalling.
• -e autorun true -e workflows true drives the same cycle for scripted am start loops.

A Test Store key is recommended (PERF_TESTER_API_KEY in local.properties): products resolve through the simulated store, no Play setup, and billing is the same in both arms. No changes to :purchases.

Known limitation: the render-readiness withTimeout can't actually bound awaitGetWorkflow/awaitGetUiConfig, because those SDK APIs use non-cancellable suspendCoroutine. A hung fetch on the workflows arm keeps the cycle waiting past the guard. The getOfferings measurement itself is properly bounded. Left as-is for now, a proper fix needs a detached-async workaround.

Verification

• :test-apps:perftester:testDebugUnitTest (3/3)
• :test-apps:perftester:assembleDebug
• detektAll
• on-device measurement runs: needs a Test Store key. The app builds, installs, launches and renders clean on a device, but I couldn't run real measurements here.

[RevenueCat-Danger-Bot]

	1 Message
📖	This PR changes 441 lines of production Kotlin/Java (limit 300); skipped via skip-pr-lines-changed-check label.

Generated by 🚫 Danger

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 80.31%. Comparing base (4766536) to head (1ed047c).
⚠️ Report is 16 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3743   +/-   ##
=======================================
  Coverage   80.31%   80.31%           
=======================================
  Files         402      402           
  Lines       16452    16452           
  Branches     2351     2351           
=======================================
  Hits        13214    13214           
  Misses       2305     2305           
  Partials      933      933           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[vegaro]

Added post-readiness memory measurements to this perftester PR. Each JSONL row now records before/after/delta RSS, RSS anon/file breakdown, and Java/native heap. The after sample is taken only after getOfferings plus the existing ui_config/workflow readiness sequence completes, so this measures retained memory after workflow loading. It is intentionally not a peak-memory metric; RSS values are read from /proc/self/status and are reported in KiB. Unit tests, assembleDebug, and detekt passed.