fix(transfer): self-dial cloned worker on read_range for SFTP/FTP pools

Clone-for-transfer workers in the SFTP and FTP pool kinds honestly
start UNCONNECTED (PD-SFTP-1 / PD-FTP-1 contract): each worker
self-dials its own connection on first transfer. download() already
calls ensure_connected() before using the session, but read_range()
did not. The new GUI segmented engine
(run_provider_segmented_download from GTC-2) puts the cloned
workers into a pool and then drives them via read_range, not
download, so the per-worker self-dial never fired and every
segmented attempt failed with "Not connected to server".

Fix: lift the same self.ensure_connected().await? guard from
download() into read_range() on both SFTP and FTP. The call is a
no-op when already connected, so the CLI pget_segmented_download
path (which pre-connects each worker via create_and_connect) is
unaffected, and the existing download() callers keep behaving
identically.

Surfaced by live-WAN validation of GTC-1 / GTC-2 against the
axpbuntu lab (see tests/gtc/reports/20260520T113538Z_gui_wan/).
After the fix, SFTP segmented byte-identical at 2.52x, FTP at
1.71x, both inside the parity-oracle spec band.

Author: axpnet

src-tauri/src/providers/ftp.rs

@@ -1513,6 +1513,12 @@ impl StorageProvider for FtpProvider {
             )));
         }
 
+        // PD-FTP-1 mirror of the SFTP path: clone-for-transfer workers
+        // start unconnected and self-dial on first transfer. The GUI
+        // segmented engine calls `read_range` directly on the pool
+        // worker, so without this self-dial every segmented download
+        // against a clone pool fails with `Not connected`.
+        self.ensure_connected().await?;
         let stream = self.stream_mut()?;
 
         stream

src-tauri/src/providers/sftp.rs

@@ -1971,6 +1971,13 @@ impl StorageProvider for SftpProvider {
         offset: u64,
         len: u64,
     ) -> Result<Vec<u8>, ProviderError> {
+        // PD-SFTP-1: clone-for-transfer workers start unconnected and
+        // re-dial on first transfer. `download()` already does this; the
+        // segmented engine (`run_provider_segmented_download`) calls
+        // `read_range` directly on the pool worker, so the same self-dial
+        // must happen here or every segmented download against a clone
+        // pool fails with `Not connected`.
+        self.ensure_connected().await?;
         let sftp = self
             .sftp
             .as_ref()