fix: cross-message reasoning dedup to prevent OpenAI rs_* ID poisoning

Summary

Fixes #545 (immediate guardrail portion)

When a workflow stream is resumed/replayed, entire assistant messages can be duplicated with blank reasoning parts that carry the same rs_* item IDs as earlier messages. OpenAI requires item IDs to be unique across input history, so these duplicates cause follow-up turns to fail.

Problem

The existing dedupeMessageReasoning only deduplicates within a single message. But the replay bug produces cross-message duplicates — a new assistant message containing only blank reasoning parts with IDs that already appeared in a previous message.

Changes

dedupe-message-reasoning.ts

• Added dedupeCrossMessageReasoning() — processes the full message array and:
  • Tracks rs_* item IDs across all messages in order
  • Removes blank reasoning-only assistant messages whose IDs were already seen (replay artifacts)
  • Strips individual blank-text reasoning parts from mixed messages
  • Preserves non-blank multi-summary segments (same ID, different text)
  • Returns the same array reference when no dedup is needed

chat.ts (workflow)

• convertMessages now applies cross-message dedup as a second pass after the existing per-message dedup

dedupe-message-reasoning.test.ts

• 10 new test cases covering: replay removal, new ID preservation, blank part stripping, multi-summary preservation, empty message removal, Azure metadata, immutability

Test results

17 pass, 0 fail (dedupe-message-reasoning.test.ts)
28 pass, 0 fail (chat.test.ts)

All existing tests continue to pass.

Scope

This PR addresses the immediate guardrail from #545 — preventing poisoned messages from reaching the OpenAI API. The root fix (chunk-index-aware workflow resume with startIndex) is a larger change that should be handled separately.