[Windows desktop] Storage misclassifies legacy segmented rollouts as Unclassified chats

Finding: In an external report on openai/codex issue 52842, the Windows desktop app 26.1007.21434 (build 13901, prod, Windows 11 10.0.26200 x64) showed roughly 1.742 GiB of existing chat transcripts as Unclassified chats on the Storage page. This was observed for seven legacy segmented rollout files named <parent-chat-uuid>_<segment-uuid>.jsonl, which belong to two chats that still exist in the app (https://github.com/openai/codex/issues/52842). This is an external report; KFA did not reproduce it. Reported evidence: the reporter says the first session_meta record's payload.id identifies the parent chat, not the final UUID in the filename. Across 258 local .jsonl files, the installed resolver's ID mismatched on 7 files (the legacy segmented ones), while the proposed parsing mismatched on 0 (https://github.com/openai/codex/issues/52842). The reporter says read-only migration inspection reported already_paginated, and that the findings do not show loss of conversations or memories (https://github.com/openai/codex/issues/52842). Reported cause (hypothesis, not maintainer-confirmed): the reporter's read-only inspection of the installed app.asar found a resolver whose regex anchors on the final UUID before .jsonl or .jsonl.zst, so for two-UUID filenames it selects the segment ID. The reporter infers that the metadata lookup then fails and the file falls back to Unclassified (https://github.com/openai/codex/issues/52842). The reporter notes that maintainers still need to verify the full native integration, and that the bundle and function names are version-specific, not upstream locations. Proposed change (not a confirmed or released fix): extend the regex to accept an optional _<segment-uuid> after the captured parent UUID, keeping lowercase normalization, suffix handling, and undefined on nonmatch. The reporter also advises keeping the category logic and threadBytes aggregation keyed to the real parent chat (https://github.com/openai/codex/issues/52842). The reporter says a patched native app was never run, so startup, storage refresh, update, and rollback behavior are untested. The expected move of 1,870,794,685 bytes from Unclassified to Chats is an expectation, not a measurement. Status: the issue is open. The supplied content shows no maintainer response, fix commit, or released fix. The only comment is from an unaffiliated account and links a sanitized ZIP of a reproducer, which was not inspected here (https://github.com/openai/codex/issues/52842#issuecomment-6096497655). No workaround was validated; the reporter explicitly declined to modify the cache category because it could leave per-chat accounting attached to a nonexistent chat. Limitations: single external report, version-specific, comments coverage may omit middle comments, and no KFA reproduction. Treat as a classification/accounting symptom only, not evidence of data loss.