#7137: fix: add openai-codex/gpt-5.2 to XHIGH_MODEL_REFS

## Summary The model ref `openai-codex/gpt-5.2` was missing from the xhigh allowlist, causing sessions with `thinking='xhigh'` to hang with 0 tokens when using this model configuration. ## Problem The OpenAI `gpt-5.2` model supports `reasoning.effort: xhigh`, but OpenClaw only had `openai-codex/gpt-5.2-codex` in `XHIGH_MODEL_REFS`. Users with custom provider configs using `gpt-5.2` (without `-codex` suffix) were unable to use xhigh thinking. ## Fix Add `openai-codex/gpt-5.2` to `XHIGH_MODEL_REFS` array. ## Testing - [x] Tested locally - xhigh spawn now works (6972 tokens vs 0 tokens before) - [x] Unit tests pass (`pnpm test -- src/auto-reply/thinking.test.ts`) ## Local Validation - `pnpm build` ✅ - `pnpm check` ✅ - `pnpm test -- src/auto-reply/thinking.test.ts` ✅ ## AI-Assisted Yes - Claude helped identify and fix the bug. <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR updates the `src/auto-reply/thinking.ts` xhigh allowlist by adding the model ref `openai-codex/gpt-5.2` alongside the existing `openai/gpt-5.2` and `openai-codex/gpt-5.2-codex` entries. This ensures `supportsXHighThinking()` recognizes `gpt-5.2` under the `openai-codex` provider (without the `-codex` suffix), so `listThinkingLevels()` can expose the `xhigh` option for that configuration instead of silently omitting it (which reportedly led to 0-token/hanging sessions). <h3>Confidence Score: 5/5</h3> - This PR is safe to merge with minimal risk. - The change is a single additive allowlist entry in `XHIGH_MODEL_REFS`, and the surrounding logic already normalizes/compares these refs via Sets; there are no behavioral changes beyond enabling xhigh for the intended provider/model combination. - No files require special attention <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->