fix(node): avoid classifying filesystem EBUSY as SQLite busy

[Dailin521]

Summary

Prevent filesystem EBUSY failures from being misclassified as SQLite busy/locked errors in the Node CLI.

Root cause

isSqliteBusyError() used a generic message.includes("busy") check. Node filesystem errors such as:

EBUSY: resource busy or locked, open '...\transaction-journal.jsonl'

therefore received the misleading state_5.sqlite is currently in use guidance.

Changes

• Match explicit SQLITE_BUSY* and SQLITE_LOCKED* codes.
• Retain SQLite's database is locked and database table is locked messages.
• Preserve unrelated filesystem EBUSY errors unchanged.
• Add focused regression tests for both filesystem and SQLite cases.

User impact

Journal/file sharing failures now retain their real filesystem diagnostic instead of telling users to close Codex because SQLite is busy.

Validation

• node --test test/sqlite-error-classification.test.js
• npm test — 210 passed

Non-goals and follow-up draft

This PR intentionally does not change rollback or recovery behavior.

A separate follow-up should investigate and design:

1. cross-process coordination between the Node CLI and Windows GUI for one Codex Home;
2. read/write sharing or bounded retry behavior for transaction-journal.jsonl;
3. resumable, idempotent recovery for journals left in rollingBack;
4. a one-click GUI recovery action bound to the transaction backup;
5. integration tests covering a GUI status scan racing a Node journal append.

That work should remain separate because it changes concurrency and data-integrity behavior beyond this error-classification fix.