[rigel-mintaka]

Summary

Owner: go-migration. Round-3 review fixes for the M4 refresher (RIG-8122). #63 merged before these landed.

• refresh relabels an error as interrupted by <signal> only when its chain carries the cancellation (context.Canceled or the SIGINT/SIGTERM cause). A real build failure that races a cancel keeps its own message.
• When an interrupt also hits a restore failure, both errors are reported; the restore error is no longer dropped.
• The early-interrupt test now asserts that no write happens, instead of a pin check that was trivially true.
• renovatehashes_test.go is split into fixture_test.go and interrupt_test.go; every file is under 500 lines.

Verification

• moon run :ci --force: all 40 tasks pass.
• go test -race ./internal/renovatehashes/... passes, including the new cases for build failure after cancel and interrupt plus restore failure.
• golangci-lint (repo config) reports 0 issues; funlen 60/40 and gocognit 20 on the package report 0 issues; gofmt -l is clean.

Risks

• Error-path and test changes only.

Refs RIG-8122

Co-authored-by: Matt Wilkinson matt@rigel.build

[linear-code]

RIG-8122

[rigel-mintaka]

Review rounds 5-8 (interdiffs from 884c49c): fixed canceled exec child classification (high), stringified cancel sites and double headline (mediums), and added regressions for init/fetch/log/drv-path cancel. Each one was red-checked. Remaining lows are style only (runner name, duplicated test setup), left as is. All-clear on 3710329.