perf: pipeline SFTP requests for upload/download

Implements bounded pipelined SFTP upload/download helpers and wires file transfers to use up to 64 concurrent SFTP requests.

Review follow-up included:
- cap server-advertised read/write lengths against local maxima to avoid oversized allocations from untrusted SFTP metadata
- bound the download reorder queue across both in-flight and pending out-of-order responses
- use fstat size information where available to avoid reads past EOF and validate unexpected short reads
- preserve remote download handle shutdown behavior after syncing with main
- add SFTP crate tests for chunk-size capping and in-memory pipelined upload/download behavior

Validation:
- cargo fmt --check
- cargo clippy -- -D warnings
- cargo test -p bssh-russh-sftp
- cargo test --lib --verbose (1222 passed, 9 ignored)
- cargo test --tests --verbose -- --skip integration_test
- GitHub Actions CI and CLA checks passed on 9ac61d2

Author: Yaminyam

ARCHITECTURE.md

@@ -12,7 +12,7 @@ bssh (Backend.AI SSH / Broadcast SSH) is a high-performance parallel SSH command
 - SSH jump host support (-J)
 - SSH configuration file parsing (-F)
 - Interactive PTY sessions with single/multiplex modes
-- SFTP file transfers (upload/download) with chunked streaming
+- SFTP file transfers (upload/download) with bounded pipelined streaming
 - Backend.AI cluster auto-detection
 - pdsh compatibility mode
 

Cargo.lock

@@ -20,6 +20,7 @@ tokio = { version = "1", default-features = false, features = [
     "macros",
 ] }
 tokio-util = "0.7"
+futures = { version = "0.3", default-features = false, features = ["std", "async-await"] }
 serde = { version = "1.0", features = ["derive"] }
 serde_bytes = "0.11"
 bitflags = { version = "2.9", features = ["serde"] }