feat(core): Enhance retention tracking and tool handling

[haiphucnguyen]

Summary

• Expanded analytics retention tracking to include 60, 90, 180, and 365+ launch counts.
• Improved "Maybe later" dialog flow to handle snooze state persistence reliably.
• Log tool execution failures (e.g., timeout, denial) to the chat history memory.
• Updated session initialization to utilize chat memory for session context.

Related Issue(s)

Type of Change

• 🐛 Bug fix
• ✨ New feature
• 💥 Breaking change
• 📝 Documentation
• ♻️ Refactor (no functional change)
• 🧪 Tests
• 🔧 Chore / tooling / CI

What Changed & Why

How to Test

./gradlew test

Screenshots / Recordings (if applicable)

Checklist

• I have read the CONTRIBUTING guide
• My branch is up to date with main
• I ran ./gradlew spotlessApply and ./gradlew detekt locally
• I ran ./gradlew test and all tests pass
• I added/updated tests for my changes (if applicable)
• I added/updated documentation (if applicable)
• My commits are signed off (DCO) — git commit -s
• This PR is focused on a single feature/fix

Additional Notes

[Copilot]

🟡 Changes recommended

The independently cached memory lookup can return null or the wrong instance for an existing chat client.

1 open finding

• Session memory lookup can diverge from cached ChatClient · New

What changed in this PR

Enhances retention analytics, prompt snoozing, and tool-failure memory handling.

Changes:

• Adds long-term retention milestones.
• Persists “Maybe later” snoozing immediately.
• Records denied or timed-out tool calls in session memory.

File	Description
ChatClientExtensions.kt	Adds synthetic tool results for approval failures.
AnalyticsEvent.kt	Documents expanded retention buckets.
Analytics.kt	Implements additional retention milestones.
Main.kt	Updates prompt callbacks and Clock import.
StarPromptDialog.kt	Separates skipping from final dismissal.
SessionManager.kt	Passes session memory into streaming.

🧠 Review effort: Balanced

---

Give feedback about Copilot approvals in this survey to enter a drawing for a $150 gift card.

[Copilot]

🟡 Changes recommended

Multi-tool failures remain incomplete, and the cache-size change breaks an eviction regression test.

2 open findings

• Update eviction regression fixtures for increased cache capacity · New
• Cancel all unresolved parallel tool requests on denial or timeout · New

1 resolved since last review

• Session memory lookup can diverge from cached ChatClient

🧠 Review effort: Balanced

---

Give feedback about Copilot approvals in this survey to enter a drawing for a $150 gift card.

[Copilot]

🟡 Changes recommended

Production callback ordering leaves the pending tool-request map empty when denial or timeout occurs.

3 open findings

• Populate pending tool requests before approval checks · New
• Update eviction regression fixtures for increased cache capacity
• Add regression coverage for approval timeout handling · New

1 resolved since last review

• Cancel all unresolved parallel tool requests on denial or timeout

🧠 Review effort: Balanced

---

Give feedback about Copilot approvals in this survey to enter a drawing for a $150 gift card.

[Copilot]

🔵 Needs a closer look

Tool recovery depends on LangChain4j callback and memory ordering that the focused fake-stream tests do not fully validate.

0 open findings

3 resolved since last review

• Populate pending tool requests before approval checks
• Update eviction regression fixtures for increased cache capacity
• Add regression coverage for approval timeout handling

🧠 Review effort: Balanced