fix(remote-transfer): keep file transfers running instead of timing out (#3235)

* fix(remote-transfer): keep file transfers running instead of timing out

Every remote file command inherited one 30s frontend request deadline, so a
slow-but-healthy transfer was reported as "Request timeout" while the host kept
transferring. Transfers now stream instead of buffering whole files, stage the
destination and publish it only after the bytes completed, and carry one
cancellable transfer id from the progress card to the backend so a stop reaches
the operation it displays.

Co-authored-by: OpenBitFun <318544290+bitfun-ai@users.noreply.github.com>

* fix(remote-transfer): align installer CI and cancellation paths

* fix(remote-transfer): finish cancelled upload sessions cleanly

---------

Co-authored-by: OpenBitFun <318544290+bitfun-ai@users.noreply.github.com>
Co-authored-by: GCWing <gcwing@foxmail.com>

Author: 3 people

OpenBitFun-Installer/src-tauri/Cargo.toml

@@ -15,10 +15,16 @@ path = "src/main.rs"
 
 [build-dependencies]
 tauri-build = { version = "=2.6.3", features = [] }
+# This standalone app does not commit a Cargo.lock. Keep the Tauri build-time
+# crates aligned with the patched 2.11 runtime instead of resolving 2.7/2.10.
+tauri-codegen = "=2.6.3"
 zip = "0.6"
 
 [dependencies]
 tauri = { version = "=2.11.5", features = [] }
+tauri-macros = "=2.6.3"
+tauri-plugin = "=2.6.3"
+tauri-utils = "=2.9.3"
 tauri-plugin-dialog = "=2.7.3"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"

docs/architecture/remote-workspace-transport.md

@@ -183,6 +183,29 @@ through the destination to preserve existing links and permissions. The final
 write is not an atomic transaction against other writers; interruption during
 that phase can have a partial or unknown outcome.
 
+Both providers publish an upload only after the transfer completed: the SFTP
+path streams into a same-directory `<hidden data dir>-upload-<id>.tmp` sibling
+and renames it over the destination, the container command does the same after
+its size check, and a downloaded file is staged beside its local destination
+the same way. A failed or cancelled transfer therefore leaves any previous
+destination unchanged, and staging temporaries are removed when the owning side
+can still reach the path. For directory transfers, files already published and
+directories already created remain when a later entry fails or is stopped.
+Because a published upload replaces its destination
+rather than writing through it, the result takes the staging file's mode and
+ownership, and a replaced symlink or hard link is not followed; that is what
+distinguishes a transfer from a workspace tool write above. A hard kill between
+staging and commit can leave one orphaned temporary, which the container
+commands sweep by age and the SFTP path leaves to the user.
+
+The desktop upload and download commands opt out of the shared 30-second
+controller deadline when their caller provides a progress and stop surface.
+Every SFTP request retains its own response timeout, the SSH transport drops a
+stalled connection, and the transfer ends when it completes, fails, or is
+stopped. File reads, writes, listings, and transfers without a stop surface keep
+the default controller deadline. A short deadline on a stoppable transfer only
+reported a slow-but-healthy operation as failed while the host kept working.
+
 Directory and stat records use NUL-separated fields. File names containing
 newlines or the delimiters used by older implementations remain round-trippable.
 The records are decoded only after the full byte stream is assembled; invalid