test: cover transfer-helper scan/repair/CLI and agent terminal, reaper, fsevent paths

[devin-ai-integration]

Summary

cargo llvm-cov --workspace showed the transfer-helper crate almost entirely untested (scan_linux.rs, repair_linux.rs, main.rs all at 0%) plus three thin spots in the agent. This adds unit tests only; workspace line coverage goes 76.27% → 87.83% (regions 76.06% → 87.61%).

Per-module before → after (lines):

module	before	after
transfer-helper/src/scan_linux.rs	0%	84.7%
transfer-helper/src/repair_linux.rs	0%	84.2%
transfer-helper/src/main.rs	0%	91.8%
transfer-helper/src/lib.rs	~59%	98.7%
agent/src/terminal.rs	~26%	89.4%
agent/src/vsock_server.rs	~4%	57.4%
agent/src/reaper.rs	~23%	69.7%

What the tests exercise:

• scan/repair: entry ordering and hashes, sparse extents, hard-link canonicalization, symlinks/FIFOs/sockets, xattrs, entry-count and path-length limits, source drift, repair convergence and metadata/hole restoration, device-node policy, and each transport verification failure (missing, renamed, wrong type, altered content).
• transfer-helper CLI: scan → repair round trip through a manifest file, usage errors, manifest canonicality/size/hard-link rejection, and atomic .partial replacement.
• agent terminal: serve_shell_stream over a UnixStream pair — output flows both ways, the session ends on shell exit and on client close — plus the injected TERM and the raw fd helper error paths.
• agent reaper: exec waits and the PID-1 sweep share one lock (exclusion verified), and start_pid1_reaper_if_needed is a no-op off PID 1 (an unrelated child stays waitable by its spawner).

Two non-behavioral production touches were needed:

• handle_fsevent_batch is now generic over the stream, so its framing/limit/deadline logic can be driven by tokio::io::duplex instead of a real vsock connection (unavailable off a guest):

  -async fn handle_fsevent_batch(mut stream: tokio_vsock::VsockStream, ...)
  +async fn handle_fsevent_batch<S>(mut stream: S, ...) where S: AsyncRead + AsyncWrite + Unpin

• New transfer-helper/src/test_support.rs (test-only) builds throwaway volumes. It probes whether TMPDIR supports user.* xattrs and falls back to target/dory-transfer-tests when it doesn't, and keeps volume paths short (v<pid>-<n>) because the socket tests would otherwise exceed SUN_LEN.

scripts/test.sh rust (fmt, clippy -D warnings, full workspace tests) passes. The remaining zero-coverage files are process entrypoints (agent/src/main.rs, runc-wrapper/src/main.rs); Swift/Xcode coverage was not measured since it needs macOS.

Link to Devin session: https://app.devin.ai/sessions/7817d379e16e467594e0f0fca962cf82
Requested by: @Augani

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring