Instance content import fixes

The zenmemes.com preview exposed closed-pipe notices and cascading legacy record rejections. Implement each numbered item in its own commit and update this checklist with its validation results.

• 1. Handle closed stdout pipes without PHP notices. Stop further output when a pager closes, while allowing imports, recovery, and cleanup to finish. Verified with a subprocess whose stdout reader closes early: import and publication finish, recovery journal clears, and stderr stays empty. php tests/run.php ImportInstanceCommandTest: 6/6 passed.

• 2. Preserve legacy creation timestamps without rewriting record or signature bytes. Recover dates from isolated source history, retain validated timestamp metadata in the destination, and use it on subsequent rebuilds and imports. Never install source Git settings or history in the destination. Verify legacy bootstrap dependencies, publication, repeat import, recovery, and rejection when reliable timestamp metadata is unavailable. Validation: 87/87 focused import, canonical reader, publication, read-model, private-site, and offline tests passed. Live preview recovered all 59 legacy dates; invalid entries fell from 893 to 15, with 922 importable files and four local conflicts exposed. No destination content was changed.

• 3. Report dependency root causes clearly. Distinguish absent dependencies from rejected or intentionally excluded records, group cascading failures, and retain per-record detail for review. Verify missing, excluded, invalid, signature, and transitive cases; repeat the live preview without mutating destination content. Validation: 90/90 focused regression checks passed on the final run, PHP lint, shell syntax, and diff whitespace checks passed. One earlier CLI recovery test returned an unexpected partial status, then passed alone and in the full rerun; its failure diagnostic now preserves command output for future investigation. Final live preview: 922 importable files, 1,376 duplicates, four conflicts, three blocked records, zero unsupported files, 158 intentional exclusions, and no PHP notices/warnings. The conflicts concern local root-001/reply-001 content and dates; two replies depend on excluded authority posts, and one further record depends on conflicting root-001. Default output groups these causes; --verbose retains per-file detail. No destination content was changed.

• 4. Show archive preparation progress. Report validation entry/byte counts, gzip integrity heartbeats, extraction file/byte totals, and legacy timestamp scan/recovery counts with per-phase elapsed time. Throttle intermediate output to roughly once per second, including during individual history queries. Validation: 13/13 focused archive, legacy, and CLI tests passed (including closed-pipe handling); PHP lint and diff checks passed. The live preview kept the same counts and produced no PHP warnings. Validation/integrity/extraction took about 2.5 seconds; legacy history lookup took 33.6 seconds for 59 dates, with visible progress throughout. No destination content was changed.

• 5. Parallelize legacy date recovery with at most four Git processes. Batch verification and rename-following history queries under the same total deadline and per-query output limit. Consume completed output immediately, stop outstanding workers on failure, and apply results in canonical path order. Recheck post hashes before saving metadata. Progress shows completed queries and active workers. Validation: 14/14 focused archive, legacy, and CLI checks passed; the five legacy tests passed again after the final cleanup review. The new regression compares serial/parallel metadata across more than one worker batch, a renamed post, and changed/untracked source files. On the same live archive, four workers recovered 59 dates in 13.8 seconds versus 25.0 seconds serially (about 45% less time), with all metadata bytes identical. This isolated benchmark did not change destination content. PHP lint and diff checks passed.

Scheduling, web UI, and importing source authority remain outside this work.