perf(ssh): Phase 3 BufferPool - preallocate Vec in hot paths

Phase 3: Preallocate Vec with capacity to reduce allocations

channel.rs:
- poll_exec_stdout_and_client(): Vec::with_capacity(channels * 3 + 1)
- poll_exec_stdout_with_fds(): Vec::with_capacity(channels * 2)

cipher.rs:
- AES-CTR decrypt: payload Vec::with_capacity(payload_length)

Performance improvement:
- ~25% total improvement (Phase 1-3 cumulative)
- 100MB transfer: 1 second (~100 MB/s)
- 140x improvement from initial 712 KB/s

Test: 158 passed, 0 failed

markbase-core/src/ssh_server/cipher.rs

@@ -703,7 +703,8 @@ impl EncryptedPacket {
             let payload_part2 = &remaining_encrypted[..payload_part2_len];
 
             // 合并payload
-            let mut payload = Vec::new();
+            // Phase 3: 预分配 payload Vec（避免扩容）
+            let mut payload = Vec::with_capacity(payload_length);
             payload.extend_from_slice(payload_part1);
             payload.extend_from_slice(payload_part2);