What changed (user-visible)

• Unreachable target: sanssh file cp through the proxy to a target that doesn't exist (or is unreachable) now fails with a clear error instead of hanging indefinitely.
• Target dies mid-upload: if the target server crashes or is killed during a file upload, the client now reports an error and exits with a non-zero code instead of silently reporting 100% success.
• Progress bar on error: the progress bar is cleared when an upload fails, so users no longer see a confusing "100%" line followed by an error message.

Why it was broken

TargetStream.Send() — the proxy method that enqueues data for forwarding to the target — only checked the gRPC stream's context for cancellation. Two problems:

1. Before connection: the gRPC stream context belonged to a placeholder (unconnectedClientStream) and was never cancelled when NewStream failed. Send() blocked forever on the full reqChan.

2. After connection: when the target died mid-stream, grpcStream.SendMsg returned io.EOF but the send loop exited without cancelling anything. Send() kept accepting data into the dead reqChan because the buffer had room (consumer gone) and the context wasn't cancelled yet.

What was done

proxy/server/target.go:

• Send() now selects on both s.ctx (our own context, cancelled by cancelFunc) and getStream().Context() (gRPC transport-level cancellation). Both are needed because after setStream() they are different objects.
• Run() calls s.cancelFunc() when NewStream fails (was missing).
• Run() calls s.cancelFunc() when SendMsg returns io.EOF (target died mid-stream), so Send() fails immediately on the next call instead of silently enqueueing data.

services/localfile/client/client.go:

• Progress bar is cleared on error paths so it doesn't render 100% after a failure.

proxy/server/target_test.go:

• TestSendUnblocksWhenTargetUnreachable — verifies Send unblocks when the target is unreachable.
• TestSendFailsAfterTargetDiesMidStream — verifies Send returns an error after the target dies mid-stream.