Respect plan_mode_reasoning_effort in Plan mode

[jason-rl]

Summary

Please have codex-acp honor the effective plan_mode_reasoning_effort from Codex configuration when entering Plan mode, and report that effective effort through ACP session config options so clients such as Zed display the value actually being used.

For example:

model_reasoning_effort = "medium"
plan_mode_reasoning_effort = "high"

If the session's ordinary reasoning effort is changed to Low and then Plan mode is selected, Codex's TUI switches to High. codex-acp currently constructs the Plan collaboration settings from the session's ordinary model/effort, so the Plan turn still uses Low and Zed continues to display Low.

Expected behavior:

• Entering Plan mode uses High and returns High as the current reasoning_effort ACP config option.
• Returning to Default mode restores the session's ordinary Low effort.
• This works whether Plan mode is selected through the ACP config option or /plan.

PR #337 persists the effort already selected by the adapter, but its current implementation still constructs Plan mode from currentModelId and does not read plan_mode_reasoning_effort, so it does not appear to cover this behavior.

Related OpenAI Codex report: openai/codex#30564

---

_{Written with OpenAI Codex and reviewed before posting.}