#7767: fix: enable turn validation for all non-OpenAI providers to prevent role ordering errors

## #7693 Third-party OpenAI-compatible providers (e.g., Kimi/Moonshot, DashScope/Qwen) configured with `api: "openai-completions"` were encountering API errors: - `"roles must alternate between 'user' and 'assistant'"` - `"incorrect role information"` These errors resulted in the user-facing message: **"Message ordering conflict - please try again."** ## Root Cause In [agents/transcript-policy.ts](cci:7://file:///c:/Users/reddragon/Documents/open-source/openclaw/src/agents/transcript-policy.ts:0:0-0:0), the [validateAnthropicTurns](cci:1://file:///c:/Users/reddragon/Documents/open-source/openclaw/src/agents/pi-embedded-helpers/turns.ts:76:0-118:1) flag was only enabled for Anthropic providers: ```typescript validateAnthropicTurns: !isOpenAi && isAnthropic, <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR changes transcript policy resolution so `validateAnthropicTurns` is enabled for all non-OpenAI providers (previously only non-OpenAI Anthropic), aiming to prevent role-ordering errors from OpenAI-compatible third-party providers. The change affects how session history is sanitized/validated before use in the embedded runner (both run and compaction flows) via `resolveTranscriptPolicy` → `sanitizeSessionHistory` → optional `validateGeminiTurns`/`validateAnthropicTurns`. <h3>Confidence Score: 4/5</h3> - Likely safe to merge, but the renamed semantics of the validation flag deserve a second look. - The PR is a small, localized change that broadens when a validation step runs; main risk is behavioral mismatch/overreach (applying an Anthropic-specific fixer to other providers) rather than correctness regressions across the codebase. - src/agents/transcript-policy.ts <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->