fix: report a clear cli error when no herdr server is running

[season179]

Current behavior

With no herdr server running, every socket CLI command fails with a raw debug string:

$ herdr workspace create --cwd /tmp --label test
Error: Os { code: 2, kind: NotFound, message: "No such file or directory" }

It reads like the --cwd path is invalid, which is exactly how #1941 was reported. A stale socket file (ConnectionRefused) and every other workspace/tab/pane/agent/worktree command produce the same raw output.

What this does

Dead-socket connect failures from the guarded CLI request path (ErrorKind::NotFound | ConnectionRefused — kind-based so Windows named pipes classify the same) map to a marker error carrying a server_not_running ErrorResponse, mirroring the existing protocol_guard pattern. The edge that surfaces the error prints it exactly once:

$ herdr workspace create --cwd /tmp --label test
{"id":"cli:workspace:create","error":{"code":"server_not_running","message":"no herdr server is running at /Users/me/.config/herdr/herdr.sock; run `herdr` to start one"}}

Exit codes are unchanged (server errors 1, syntax errors 2). Printing is deferred — the marker carries the response — so callers that intentionally recover from a dead server keep their behavior with nothing printed:

• plugin commands still fall back to the offline registry (plugin list --json → offline result, exit 0, empty stderr)
• herdr status still reports status: not running
• agent start transport failures print exactly one error line
• session stop messaging unchanged

The shared classifier replaces the private copy in cli/status.rs. No wire protocol change (CLI-emitted error, like protocol_mismatch) and no server auto-start behavior.

Validation

• just check green on macOS (zig 0.15.2 via homebrew) except one pre-existing failure unrelated to this change: live_handoff_keeps_unmanaged_agent_name_bound_to_saved_session fails identically on clean master @ 1491b7d
• two focused regression tests: dead-server mapping (request id, code, socket path in message, marker recognizable without string matching) and a classifier negative case
• manually verified: dead socket → single JSON error + exit 1 across command groups; status graceful; positive path against a live server unchanged; plugin offline fallback behavior identical to v0.7.5

refs #1941

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 5c222dc5-0fc1-44c9-bff7-efa82ac1bc9d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Greptile Summary

Maps dead-socket CLI connect failures to a deferred server_not_running JSON error while preserving recover paths.

• Shared kind-based classifier and ServerNotRunningReported marker in server_not_running / cli.rs
• send_request and send_request_unchecked both map NotFound/ConnectionRefused to the marker
• main and agent transport print the deferred response once; plugin offline fallback still treats the marker as a connection error
• Status reuses the shared classifier; focused unit tests cover mapping and negative kinds

Confidence Score: 5/5

The PR appears safe to merge; the prior unchecked-path gap is fixed and no blocking failure remains.

send_request_unchecked now applies the same dead-server mapping as send_request, so callers get the deferred server_not_running marker; recover paths and single-print edges remain consistent.

Important Files Changed

Filename	Overview
src/cli.rs	Friendly dead-server mapping on both send_request and send_request_unchecked, plus classifier helpers and tests.
src/cli/server_not_running.rs	Deferred marker error carrying ErrorResponse, mirroring protocol_guard style without eager print.
src/main.rs	Prints carried server_not_running JSON once and exits 1 when the marker reaches main.
src/cli/agent.rs	Surfaces deferred server_not_running response instead of a second generic transport line.
src/cli/plugin.rs	Treats the marker as a connection error so offline registry fallback still prints nothing.
src/cli/status.rs	Uses shared server_not_running_error classifier; not-running status path unchanged in behavior.

_{Reviews (2): Last reviewed commit: "fix: map dead-socket errors on the unche..." | Re-trigger Greptile}