fix(assistant): harden conversation identity and pre-SSE errors

[vitorhugo-dotnet]

Summary

Companion frontend fix for vitorhugo-dotnet/SpringBoot-JobApplyTracker#96.

Root cause / production investigation

• production Cloudflare Worker deployment is sourced from current main
• commit 7657295 (PR fix(chat): complete conversationId handling for assistant SSE #117) finished deploying at ~21:50 BRT
• Loki trace 5121fb9b824c7cfd315df13473b9544d failed at 21:54:10 BRT, after that deploy
• the pre-feat(assistant): add failed-message retry and rate-limit countdown #115 client (c4fe2fe) sent JSON.stringify({ message }) with no conversationId at all
• therefore the production failure is consistent with an already-open stale client/tab still running the pre-feat(assistant): add failed-message retry and rate-limit countdown #115 bundle; the backend deserialized the missing UUID as null
• there is only one current /assistant/chat request path in the repository
• the app is a PWA with precached JS/HTML, so this stale-client transition is technically possible even though the Worker itself was current

Changes

• centralize assistant conversation UUID creation/validation
• revalidate and self-heal the conversation ID immediately before every request
• recover from IndexedDB/open/read/decryption failures with a fresh UUID
• keep replacement IDs persisted through assistant history
• send Accept: text/event-stream, application/problem+json, application/json
• parse structured non-2xx JSON/problem responses before touching the SSE parser
• reject successful non-SSE responses as typed AssistantRequestError
• preserve conversation ID on retry and generate a fresh ID on Clear
• add Playwright regression coverage for request headers, IndexedDB failure recovery, Clear identity rotation, and JSON-vs-SSE handling

Backend issue: vitorhugo-dotnet/SpringBoot-JobApplyTracker#96