fix(client): don't reject unexpected success when no 2xx is declared

[iamralch]

Fixes the first half of #89: the duplicated status.is_success() branch. The variant naming it also mentions (AmberStrict_2) is left for a separate PR.

What was wrong

An operation that declares no 2xx response, such as one with only a default response, selects every successful status, so its success guard is status.is_success() itself. The branch after it, which rejects successful statuses the return type didn't select, tested the same condition:

if status.is_success() {
    Ok(())
} else if status.is_success() {
    Err(/* unexpected successful status */)
} else {
    // ...
}

It couldn't be reached, and clippy's deny-by-default ifs_same_cond rejected the generated client.

The change

generate_error_handling emits the branch only when there are selected statuses to reject (!success.statuses.is_empty()). That covers the buffered path, which is the one clippy caught, and the binary and streaming paths, where the same check was nested inside the else and just as unreachable.

Tests

• tests/default_only_response_test.rs: a default-only operation tests status.is_success() once and has no unexpected-success branch, and an operation with a declared 204 still rejects other successes. It fails before this change.
• cargo test --all-features: 719 passed, none failed.
• tests/corpus-manifest.txt is regenerated. scripts/gen-diff.sh upstream/main shows 15 specs moved, all in client.rs, by 342 removed branches. For cloudflare (43), github (8) and google-youtube (83), removing exactly those branches from the base output gives the new output byte for byte. The +141 lines in the Cloudflare diff are git aligning similar functions differently.

cargo clippy --all-features -- -D warnings reports one nonminimal_bool in src/schema_roundtrip.rs with clippy 0.1.93. It's on main as well, so it isn't from this change.

[vercel]

@iamralch is attempting to deploy a commit to the lbl-rd Team on Vercel.

A member of the Team first needs to authorize it.