LLM-24555 Update codex to 0.99.0

[slapoguzov]

• Rate limits are now per-tier. The /status command displays each rate limit separately (e.g. "Standard 1h limit: 75% left", "Fast 1d limit: 20% left") instead of a single limit.
• Image input is rejected for text-only models. If the active model does not support image input, prompts containing images are rejected with an error.
• Reasoning summaries are disabled for non-reasoning models. Models that do not support reasoning no longer request reasoning summaries.
• Context compaction is now visible. When the conversation is compacted to fit the model's context window, a message "Context compacted to fit the model's context window." is shown in the chat.
• Plan type shown in status. The /status command now displays the account plan type (e.g. "ChatGPT pro (user@example.com)").
• Minimum Codex version bumped to v0.99.0.