cv4pve-node-protect: all versions and old releases

Added
- Documentation site: corsinvest.github.io/cv4pve-node-protect with getting started, SSH access and security, connection, what to back up (and why /etc/. alone misses /etc/pve), archives and retention, scheduling, a step-by-step restore guide and the .NET library. Every page was checked against the code and a real backup (#12, #16)
- Library: ProtectHelper with host parsing, archive names and retention as the command line tool does them (ParseHosts, ParseHostAndPort, GetArchiveFileName, IsBackupDirectoryName, GetBackupsToDelete), for callers that want the same layout (#15)
- Unit tests (tests/Corsinvest.ProxmoxVE.NodeProtect.Api.Tests), run by the CI (#15)
Changed
- A node that fails no longer stops the run: the other nodes are still backed up, each failure is printed as ERROR [node]: …, and the run ends with Backup failed for N of M node(s): … Retention not applied. and exit code 1. Retention is skipped when a node failed, so good backups are never deleted to make room for an incomplete run (#14)
- Skipped paths are reported: a path that does not exist or cannot be read is still left out, but tar's message is now printed as a warning instead of being discarded (#14)
- Archives readable only by their owner on Linux and macOS: the dated folder is created with mode 700 and the archives with 600, since they contain password hashes, keys and the cluster database (#14)
- --username is now always required, also with --private-key-file: the error is shown before anything is created (#14)
- A path containing a single quote is rejected before connecting to the node (#15)
- Updated Corsinvest.ProxmoxVE.Api.Extension and Api.Console from 9.1.15 to 9.2.4 (#11, #18)
- Updated SSH.NET from 2025.1.0 to 2026.0.0 (fixes GHSA-mggc-4xg6-vcxf and GHSA-q939-rpr3-3284 in ScpClient, which node-protect does not use) (#11)
- Product icon (Lucide shield-check), Windows executable icon and NuGet package description (#11)
- Project metadata, symbols (Source Link, .snupkg) and code style aligned with the other cv4pve tools; CI on the shared cv4pve workflow (#10, #11)
- README shortened: the details are on the documentation site (#12)
Fixed
- A path in --paths starting with - is now always a path: tar gets -- before the paths. Before, it was read as a tar option, so a value like --checkpoint-action=exec=… could run a command on the node (#17)
- Restore instructions: the README extracted archives with tar -xvzPf … -C /root/restore. With -P, tar ignores -C and writes to the absolute paths, overwriting the live files of the node. The restore guide on the documentation site extracts without -P into a staging folder (#12)
- --debug and --log-level now show the backup engine's log: the tar command run on each node and how long the transfer took. Before, they only added the stack trace to errors (#13)
- A failed run no longer leaves an empty dated folder, which counted as a backup for --keep (#14)
- IPv6 hosts on Windows: the : of the address is replaced by _ in the archive name, since Windows does not allow it in file names. Before, the run failed (#14)
What's Changed
- ci: use shared cv4pve-tools CI workflow and declare publish permissions by @franklupo in #10
- Align project metadata, symbols and icon with the cv4pve tools by @franklupo in #11
- fix: show the engine log with --debug and --log-level by @franklupo in #13
- docs: documentation site with Starlight and the shared cv4pve theme by @franklupo in #12
- fix: keep going when a node fails, report skipped paths, protect archives by @franklupo in #14
- test: unit tests, host parsing and retention moved to the library by @franklupo in #15
- docs: sync with the current behaviour of the tool by @franklupo in #16
- fix: end tar options before the paths by @franklupo in #17
- chore: SDK 9.2.4, docs theme 2.6.1, no em dashes by @franklupo in #18
- chore: release 2.2.0 by @franklupo in #19
Full Changelog: v2.1.1...v2.2.0

Changed
- Backup no longer fails when a listed path is missing on the node — tar now runs with --ignore-failed-read, so optional paths (like /root/scripts or /var/lib/ceph/. on nodes without Ceph) are silently skipped instead of aborting the whole run
What's Changed
- v2.1.1: tolerate missing paths with --ignore-failed-read by @franklupo in #9
Full Changelog: v2.1.0...v2.1.1

Added
- IPv6 support in --host — you can now target nodes by IPv6 address. Use brackets if you also specify a port: [fe80::1]:2222. Without a port, brackets are optional: fe80::1
Changed
- Backup no longer uses temporary files on the node — the archive is streamed directly from the Proxmox node into your local file. Nothing is written to /tmp on the node, so crashes or interrupted runs don't leave leftover files behind
- Hosts, paths and SSH ports that contain unusual characters are now handled safely
- Console output is shorter and easier to read: "Create config" and "Delete Backup" show the folder name instead of the full path
Fixed
- The NuGet package now publishes the Corsinvest.ProxmoxVE.NodeProtect.Api library instead of the console executable, so other projects can depend on the backup engine as a library
What's Changed
- v2.1.0: IPv6 host support, stream-based backup, new ToStream API by @franklupo in #8
Full Changelog: v2.0.1...v2.1.0

Fixed
- The NuGet package now publishes the Corsinvest.ProxmoxVE.NodeProtect.Api library instead of the console executable, so other projects can depend on the backup engine as a library
What's Changed
- v2.0.1: publish the .Api library to NuGet instead of the console exe by @franklupo in #7
Full Changelog: v2.0.0...v2.0.1

Added
- SSH private key authentication (--private-key-file, --passphrase) — log in with a private key instead of a password
- Read password from file (--password=file:/path) — keep the password out of shell history and process list
- Custom SSH port per host (--host=pve1:2222,pve2:2222)

Changed
- --paths and --directory-work are now required when running backup
- SSH port in --host=host:port is now actually honored (previously always used port 22)
- Backup retention only deletes timestamped backup folders — other folders in the work directory are left alone
- Updated to run on .NET 10 (the library still works on .NET 8, 9 and 10)

Removed
- upload command — it only copied the archive to the node without restoring anything. Manual restore via scp + tar is documented in the README.
- --dry-run option — it had no effect

Fixed
- Clear error when using --private-key-file without --username
- Temporary files on the node are always cleaned up, even if the backup fails halfway through
- Paths containing spaces or special characters are now handled safely

Security
- Temporary backup archives on the node now use restricted permissions and unpredictable names — other users on the same node can no longer read the cluster configuration while a backup is in progress