fix(go): respect build failure exit status

[jsiu93]

Closes #2185.

Summary

• Gate rtk go build success output on the wrapped process exit code.
• Surface unrecognized non-zero go build output instead of reporting Go build: Success.
• Preserve Go package-pattern/module-resolution errors such as pattern ./...: directory prefix ....

Test plan

• cargo fmt --all -- --check
• cargo clippy --all-targets -- -D warnings
• cargo test --all --quiet
• Manual repro in an empty directory: target/debug/rtk go build ./...

[CLAassistant]

All committers have signed the CLA.

[pszymkowiak]

Thanks for the thorough fix — the diagnose matches the repro in #2185 exactly, and the two-layer approach (allow-list entry for pattern + exit-aware fallback) is the right call. The runner.rs refactor (RunMode::FilteredWithExit + shared run_captured_filter) is a particularly valuable piece: it lays the plumbing so other ecosystems can opt into exit-aware filtering without duplicating logic. Several other filters (cargo clippy, mypy, ruff, tsc, eslint, golangci-lint, go vet) have the same latent "no errors detected = Success" pattern and would benefit from the same migration.

A second independent review (code-reviewer agent) and empirical sweep across 19 commands on develop vs this branch confirmed byte-identical behavior on every non-go-build path.

A few light suggestions for follow-up, none blocking:

1. ANSI escape codes in the failure fallback: format_go_build_failure echoes raw stderr lines through truncate without passing through strip_ansi. Real go build doesn't emit ANSI, so this is a theoretical concern, but a malicious or unusual go wrapper could inject OSC 8 hyperlinks or alt-screen sequences that survive into the LLM context. A strip_ansi call per line would close that gap cheaply.

2. "pattern " prefix in is_go_build_error_line: matches any line starting with pattern (lowercased). A custom build tool printing pattern ./internal/...: 12 files matched would be classified as an error. Tightening to "pattern ./" or requiring ": " after the argument would eliminate that risk while still matching every real go build pattern-error shape.

3. Tests: the new tests use assert!(result.contains(...)) — consistent with the rest of go_cmd.rs, so just a style note. Adding edge-case tests for filter_go_build_with_exit("", 1) and whitespace-only inputs would round out coverage.

4. Follow-up opportunity (out of scope for this PR): migrating cargo clippy, mypy, ruff, tsc, eslint, golangci-lint, and go vet to run_filtered_with_exit would close the same class of bug across the rest of the codebase.

Approving — the fix is correct, the refactor is well-bounded, and behavior is preserved across every other command. Thanks for the careful work!