fix: address review feedback for calls feature

[noanflaherty]

Summary

• Gate answer persistence on orchestrator waiting_on_user state before marking question as answered (ci: add web and platform CI/PR workflows #4)
• Terminate Twilio call via provider API before marking session completed in call_end tool (Create top level setup.sh script #5)
• Mark call session as failed with error details when call initiation throws (docs: update setup instructions and environment configuration #9)

Verified

• Issues feat: initialize Next.js app in /web directory #1 (control tag filtering), feat: add platform terraform for GKE deployment #2 (Twilio auth bypass), ci: add terraform apply workflow on platform changes #3 (CallSid capture), Replace all references of GCP_REGION with us-central-1 #6 (deterministic ordering), feat: add top-level setup.sh script #7 (pending status enforcement), Upon opening home page, don't auto redirect to creating the assistant #8 (status reset on user answer) were already fixed in prior PRs

Test plan

• call-orchestrator tests pass (8/8)
• call-state tests pass (10/10)
• call-store tests pass (23/23)
• No new TypeScript errors introduced

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 5 additional findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bcb5bf1e23

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid blocking teardown on provider hangup call

This await provider.endCall(...) runs before relay/orchestrator cleanup, so call termination now depends on Twilio responsiveness; when Twilio/network is slow or stalls, call_end can hang and never reach the local cleanup path, leaving the active relay/orchestrator alive instead of ending promptly. Because endCall uses a plain fetch without an abort timeout, this regression can surface during provider incidents and prevents the tool from reliably stopping calls.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Don’t mark session failed after post-initiation errors

The new catch-path update applies to any exception after sessionId is assigned, including failures that happen after Twilio has already accepted the outbound call (for example, an error persisting providerCallSid), so we can persist status: 'failed'/endedAt while the call is actually live. That corrupts lifecycle state and can hide an active call from getActiveCallSessionForConversation, so this failed transition should be limited to genuine initiation failures.

Useful? React with 👍 / 👎.