LRGEX-Compress Files

🛡️ Data Integrity — Full Metadata Preservation

LRGEX Compress now preserves your files exactly across a compress → extract round-trip. Whatever goes into the archive comes out identical — no more "every extracted file shows today's date."

• LastWriteTime (mtime) preserved on every file and folder — sort-by-date, build systems, sync tools, and incremental backups all keep working correctly.
• CreationTime (ctime) preserved — the original creation date is restored, not lost.
• Windows attributes preserved — Hidden, Read-only, System, and Archive flags survive the round-trip. A hidden file extracts hidden; a read-only file extracts read-only.
• Symbolic links / junctions preserved — links recreate as links. On Windows, creating a link needs Administrator or Developer Mode; LRGEX asks once ("allow admin?") and, if yes, re-launches the extraction elevated to recreate every link exactly. If you decline, links degrade gracefully (never aborts the whole extraction).
• Directory timestamps restored in a final pass — so writing children into a folder doesn't overwrite the folder's original mtime. NTFS alternate data streams are not supported (rarely used; not planned).

⚡ Performance — Extract & Compress Wins

• Metadata sidecar (.lrgex/meta.bin): one small packed blob at the front of the archive replaces 11k+ per-entry metadata headers on big folders. Both compress and extract handle roughly half as many tar entries now. Old v1.4.0 archives still read correctly via the per-entry fallback.
• One-syscall metadata restore (SetFileInformationByHandle): mtime + ctime + attributes set in a single call on the already-open file handle — no per-file re-open.
• Direct 4 MB chunked writes on large-file extract (no redundant BufWriter copy).
• Cache-served metadata on compress walk (DirEntry::metadata() instead of a fresh CreateFile per file) — saves thousands of syscalls on big folders.
• Removed the redundant 4 MB BufWriter on compress output — zstd already emits good-sized blocks; the buffer was a pure memcpy of every compressed byte.
• Tuned zstd multithreading — JobSize(16MB) + OverlapSizeLog(0) at level 1, fewer thread handoffs.
• Dedicated oversubscribed write pool (3× core count) for extraction writes — correct sizing for workers that block in CreateFile and Defender scans.

🐞 Fixes

• Multi-select coordinator hardened: handles WAIT_ABANDONED (previous instance crashed holding the mutex) and stale coordinator-PID lockfiles — a second rapid launch can no longer silently no-op.
• Round-trip verified lossless: mtime/ctime/Hidden/ReadOnly/System, Unicode names, deeply nested paths, and a 3000-file stress batch (SHA256-checked).

🔒 Security Hardening

• Signed manifest (detached latest.json.sig): the auto-update manifest is now cryptographically signed with Ed25519, preventing a compromised server from injecting a fake version or download URL.
• Download size cap (100 MB): the updater rejects files exceeding 100 MB, preventing memory exhaustion from a poisoned manifest pointing at a huge file.
• Update-channel logging: failed update checks now write to update.log in the install folder, so silent update-channel failures are observable.
• GitHub release asset verification: deploy.ps1 now size-verifies the GitHub release asset via the API, not just the exit code.
• Path-length guard: right-clicking a file with a path longer than 247 characters shows a clear error instead of a confusing "not found".
• Slint attribution: added "Made with Slint" badge (Slint Royalty-free License §2).