feat(cli): give failures a taxonomy exit code instead of a blanket 1

[joshrotenberg]

Part of #1045, item 3 (GAP-AUTOMATION-02).

Problem

Every runtime failure exited 1, from a blanket process::exit(1). A script could tell that redisctl failed but not why, so branching on failure type meant matching on message text.

Change

crates/redisctl/src/error.rs gains an exit_code module holding the taxonomy the issue names, and RedisCtlError::exit_code() mapping each variant to one:

Code	Name	Variants
1	GENERIC	Other, OutputError
2	USAGE	UnsupportedDeploymentType, FileError
3	CONFIG	Configuration, ProfileNotFound, ProfileTypeMismatch, NoProfileConfigured, MissingCredentials
4	AUTH	AuthenticationFailed
5	NOT_FOUND	ApiError (404)
6	VALIDATION	InvalidInput
7	CONFLICT	ApiError (409)
8	UPSTREAM	ApiError (other)
9	RATE_LIMITED	ApiError (429)
10	NETWORK	ConnectionError
11	TIMEOUT	Timeout
12	CANCELLED	Cancelled

The match is exhaustive, so a new variant has to declare its code, the same way code() already works.

main.rs exits with e.exit_code() rather than 1. The two pre-dispatch exits in the prefix-inference layer do not have a RedisCtlError in hand, so they name their code directly: the ConfigError arms exit CONFIG, and the "cannot determine platform" arm exits USAGE.

The -o json/-o yaml envelope now carries exit_code alongside code. That is the field #1048 explicitly deferred to this change.

Choices worth flagging

• USAGE is 2 because clap already exits 2 when argument parsing fails. Both mean the invocation was wrong, so a script sees one code for that class rather than two.
• GENERIC stays 1, so anything not yet classified keeps its current behavior. Other is the anyhow catch-all and OutputError covers serialization and IO.
• ApiError sniffs the status out of its message. It is the only variant that carries a server status, so without this, not-found, conflict and rate-limited would be unreachable codes. suggestions() already does the same for 404. The real fix is a status field on the variant, which belongs with the Config/Configuration merge in item 2 of Automation contract: tracing corrupts -o json, duplicate error line, exit codes, error envelope #1045.

Gap found while testing

--profile <nonexistent> still exits 1, not 3. It never reaches RedisCtlError::ProfileNotFound: connection.rs:174 and connection.rs:389 report it through anyhow::Context::with_context, which lands in Other. This is the "354 error paths collapse into a generic variant" problem described in item 2 of #1045, and fixing it means changing connection.rs, which is outside this change. Worth doing next, since a nonexistent profile is the most common config failure there is.

Behavior

Exit codes only. No message, format or control-flow changes. Success still exits 0.

$ redisctl enterprise database list; echo $?
error: Configuration error: No enterprise profiles configured
3

Checks

cargo fmt --all -- --check, cargo clippy --all-targets -- -D warnings and cargo test all pass locally (full suite green). cargo doc --no-deps --all-features produces only the pre-existing warnings in cli/enterprise.rs and redisctl-mcp, unchanged by this diff.

Tests added: four unit tests in error.rs covering the per-variant mapping, the ApiError status split, the envelope carrying the code, and an assertion that no classified variant falls back to 1. The existing -o json envelope test now also asserts the process exit code matches the envelope, plus one integration test that a config failure exits 3.