fix(daemon): promote every error field alias so control-listener failures are diagnosable

[posthog]

Problem

• When the control socket fails to bind, control_listener_loop_actor returns an error and its wrapper calls request_shutdown(), so the whole daemon stops and the user loses all tracking. Whoever hits it gets no tracking at all.
• The captured $exception is undiagnosable: it arrives as a bare static string, with no cause and no stack trace. You cannot tell whether the bind failed on a stale socket, a permissions problem, or something else.
• Root cause: message_with_promoted_error in sentry_layer.rs appended the underlying cause to the exception value only when the tracing field was named literally error. The %e call sites name the field e, so the AutterError text stayed in the context map and never reached the exception value.
• Blast radius: the field-name gap affects about a dozen tracing::error!(%e, ...) sites in src/daemon.rs, including the sibling trace listener. Every one lands as an undiagnosable bare message.

Changes

• Promote every alias. message_with_promoted_error now promotes the first non-empty of error, err, e, or source, so all spellings reach the exception value and distinct causes fingerprint separately in error tracking.
• Name the setup failures. The two control-socket setup steps that propagated bare io errors — stale socket removal and owner-only permissions — now name themselves, matching the already-named bind step. Each real cause becomes its own issue.

Setup step	Before	After
stale socket removal	IO error: <os error>	failed removing stale socket <path>: <os error>
bind	failed binding control socket: ...	unchanged
owner-only permissions	IO error: <os error>	failed setting owner-only permissions on socket <path>: <os error>

Expected impact

• Diagnosability only; no behavior change. The failure rate should stay flat.
• The single bare issue splits into one issue per real cause, which is what tells us whether this is a permissions problem, a stale socket, or something else.

Tests

• cargo test --lib sentry_layer — 8 pass, including new cases for the e, err, and source aliases and for skipping an empty earlier alias.
• cargo check --locked, cargo fmt --check, and cargo clippy --lib all clean.

Agent context

• The inbox report named the repository as autter-dev/autter-runtime, but that is the TypeScript/Node SDK monorepo. The Rust CLI and daemon live in autter-dev/autter-cli, so this fix targets autter-cli.

---

Created with PostHog Desktop from this inbox report.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}