Include archived Codex sessions in usage reports (rebased, supersedes #663)

[ozymandiashh]

Rebases #663 by @gt-zhangshihao03 onto current main and opens it as a maintainer PR. The original had a merge conflict with main after #656 landed; the code is unchanged from #663 apart from resolving a README provider-table conflict. The original commit authorship is preserved.

What it does

Discovers Codex rollout-*.jsonl files from the flat ~/.codex/archived_sessions directory in addition to the date-partitioned ~/.codex/sessions/YYYY/MM/DD, using the same metadata validation and project attribution for both. Archiving a Codex conversation (which moves its rollout file to archived_sessions) no longer erases its historical usage from reports.

Maintainer verification

• No double-counting if a session were ever present in both directories: parsing dedups on a content-based key (codex:${forkedFromId||sessionId}:${cumulativeTotal}:…) via a seenKeys set created once and shared across all sources (src/parser.ts). Identical content at two paths yields identical keys, so the second copy is skipped. Confirmed on code.
• Archived-only home works: the readdir(sessionsDir).catch(() => []) change lets a home with no sessions/ directory fall through to the archived scan. The regression test builds a fresh temp home containing only archived_sessions/, so it genuinely exercises that path (the old early-return would have failed it).
• Ordering: archived sources are appended after active ones; this doesn't affect totals or time-series because the day-aggregator buckets each session by its own firstTimestamp and sorts, rather than relying on discovery order.
• npx tsc --noEmit clean; npx vitest run tests/providers/codex.test.ts green (22 tests).

Fixes #662
Supersedes #663 (original work by @gt-zhangshihao03)