fix: read model provider from config on session resume

[xiatg]

Summary

• Reverts PR fix: explicitly pass model provider on session resume #139 which introduced getResumeModelProvider() with a hardcoded fallback to "openai"
• The previous change forced the model provider to default to "openai" when none was explicitly passed, overriding the user's configured provider in ~/.codex/config.toml
• Per the official configuration guide, users can set their model provider in ~/.codex/config.toml — the hardcoded fallback bypasses this

🤖 Generated with Claude Code

[AlexandrSuhinin]

Proper fix should set provider from the config then. Reverting this PR will break resume on sessions if provider was changed.

[xiatg]

Updated to use getCurrentModelProvider() in getResumeModelProvider(), which reads model_provider from the config as a fallback instead of only checking the environment variable. The "openai" fallback is kept for the case where no provider is configured anywhere.