What does this PR do?

Fixes several Go CI Visibility edge cases found while reviewing retry, cleanup, coverage, and Git metadata handling after #4717 and #4722.

This PR updates Go test retry orchestration so cleanup failures behave like normal test failures. A cleanup panic is recorded on the failing attempt, cleanup Fatal / FailNow no longer escapes the retry loop through runtime.Goexit, and cleanups now run after Datadog-managed parallel subtests have completed so parent cleanup order remains consistent with Go's testing lifecycle.

It also fixes Orchestrion subtest panic handling when a Datadog retry wrapper owns the execution. In that case, subtest panics are recorded as failed attempts and returned to the retry owner, instead of immediately repanicking before retries can run.

Other fixes included here:

• stop automatic flaky retries when the global retry budget reaches zero
• protect benchmark instrumentation map writes with the write lock
• synchronize coverage writer payload rotation between add and flush
• report test.code_coverage.enabled=false when ITR is enabled but coverage is not actually active
• avoid an index panic when Git reports a pack hash but the expected .pack file is missing

Motivation

The previous CI Visibility retry fixes addressed known lifecycle bugs, but a follow-up review found more cases where retry attempts could be misclassified, skipped, or allowed to race with shared state.

The highest-risk issues were in Go cleanup and Orchestrion subtest panic flows. Both are part of the real test result, so they need to pass through the same retry owner that handles failures in the test body. Without that, a cleanup failure could be hidden or abort retry orchestration, and a retryable subtest panic could fail the process before Datadog could schedule the next attempt.

The smaller fixes remove incorrect state reporting and race-prone code paths found during the same review.

Testing

• go test ./internal/civisibility/integrations/gotesting/failnow -count=1
• go test ./internal/civisibility/integrations/gotesting -count=1
• go test ./internal/civisibility/integrations/gotesting/subtests -count=1
• go test ./internal/civisibility/utils -run 'TestCreatePackFiles|TestGit|TestPackFiles' -count=1
• go test -race ./internal/civisibility/integrations/gotesting/coverage -count=1
• Bypass=true go test -race ./internal/civisibility/integrations/gotesting -run TestGetFieldPointerFrom -count=1
• PATH="$(go env GOPATH)/bin:$PATH" GOFLAGS="-timeout=30m -tags=githubci,buildtag" go run github.com/DataDog/orchestrion go test -count=1 ./parallel_testing_retries
• git diff --check main..HEAD

Reviewer's Checklist

• Changed code has unit tests for its functionality at or near 100% coverage.
• System-Tests covering this feature have been added and enabled with the va.b.c-dev version tag.
• There is a benchmark for any new code, or changes to existing code.
• If this interacts with the agent in a new way, a system test has been added.
• New code is free of linting errors. You can check this by running make lint locally.
• New code doesn't break existing tests. You can check this by running make test locally.
• Add an appropriate team label so this PR gets put in the right place for the release notes.
• All generated files are up to date. You can check this by running make generate locally.
• Non-trivial go.mod changes, e.g. adding new modules, are reviewed by @DataDog/dd-trace-go-guild. Make sure all nested modules are up to date by running make fix-modules locally.