What happened?

#4603 updated the openai-codex model list, but incorrectly set the context window limit for openai-codex/gpt-5.3-codex-spark to 272000 via the CODEX_CONTEXT constant.

This causes pi to trigger auto-compaction at only around 47% of the displayed context window limit when using this model.

Steps to reproduce

1. Select openai-codex/gpt-5.3-codex-spark as the model
2. See that Pi shows the context window limit as 272k
3. Fill up more than 128k tokens of the context window

Expected behavior

The context window for openai-codex/gpt-5.3-codex-spark should be 128000, not 272000 (CODEX_CONTEXT). Should just be a change on this line:

Version

0.75.5