Avoid full Codex session reads when listing and looking up sessions

[Vottivott]

Summary

This fixes two Codex-specific performance problems in ccshare:

• avoid reading entire Codex .jsonl files just to extract the first-line session_meta.cwd
• avoid doing a full listSessions() scan when the user already provided an explicit Codex session id

Changes

• add a bounded readFirstLine() helper for Codex metadata discovery
• make Codex session listing read only the first line of each file
• add a direct Codex session lookup path by exact id / unique prefix
• keep the existing fallback behavior for Claude, Gemini, and numeric index lookups

Why

Large Codex session files can be multiple GB. The previous implementation used readFile(..., 'utf-8') for every Codex session during list, and also for explicit session-id operations because getSession() always went through listSessions() first.

That makes ccshare list and ccshare share/export/preview <codex-session-id> unsafe on large histories.

Verification

Manual verification in an isolated temp HOME:

• bun run src/cli.ts list --limit 5 succeeded with a 1 GB sparse Codex session file
• bun run src/cli.ts export 019c70c7-9e1f-70c1-84a8-1405892852f1 -o out.html succeeded with an explicit Codex session id

Notes

I did not rely on tsc --noEmit for validation here because the repo currently has unrelated missing dependency / type environment issues in this environment.