What does this PR do?

This PR fixes three CI Visibility issues found while validating Test Optimization in downstream repositories:

1. It preserves mocktracer when CI Visibility starts after a package has already called mocktracer.Start().
2. It drains and closes successful CI Visibility writer HTTP response bodies so test-cycle upload connections can shut down cleanly before goleak runs.
3. It stops Early Flake Detection from retrying tests that cleanly skip on the first execution.

For the mocktracer startup path, tracer.Start() now keeps the existing civisibilitymocktracer installed as the process global tracer when CI Visibility starts later. Instead of replacing that mock-compatible wrapper, it injects the real CI Visibility tracer into it. Normal application spans keep going to the mock tracer, while CI Visibility test/session/module/suite spans go to the real tracer.

The PR also adds a finished-chunk submit hook so CI Visibility chunks are still submitted through the real tracer even though the process global tracer remains the mock wrapper. The mock wrapper now owns its real tracer delegate explicitly, protects delegate replacement and span routing with locks, keeps real-span routing state instance-local, and preserves shutdown behavior so user mocktracer.Stop() calls do not kill CI Visibility before package exit.

For the HTTP cleanup path, the CI Visibility trace writer now owns every non-nil body returned by ddTransport.send. It drains and closes those bodies on success and on the defensive (body, err) path, preserving retry behavior while allowing the HTTP transport to release keep-alive connections.

For the EFD skip path, clean skips now stop at the common EFD retry decision. That keeps pass/fail EFD retries unchanged, but avoids burning the retry budget on executions that only produce skip signal. The final-status prediction mirrors the same decision so the single skipped execution receives test.final_status=skip immediately and does not get retry tags.

No new environment variables are introduced, and this does not change /civisibility or /internal/civisibility.

Motivation

The downstream mocktracer failure happens when test packages start mocktracer in TestMain, and the test optimizer later starts CI Visibility through RunM.

Before this change, CI Visibility startup replaced the global civisibilitymocktracer with either the real *tracer.tracer or the CI Visibility noop wrapper. Code already using the v1 mocktracer path still expected the global tracer to be mock-compatible, so finishing adapted mock spans could panic with an interface conversion error.

Preserving the wrapper fixes that ownership problem without losing CI Visibility data: user test spans remain available through FinishedSpans(), and CI Visibility spans still reach the real tracer and submit to the test-cycle intake.

A later downstream run also exposed a goleak failure with net/http.(*persistConn).readLoop and net/http.(*persistConn).writeLoop after tests had passed. The CI Visibility writer was ignoring successful response bodies returned by the transport, leaving the client connection in use. Draining and closing those bodies makes shutdown deterministic for the HTTP transport.

Downstream EFD validation also showed output like skipped after 10 retries by Datadog's early flake detection. Retrying a clean skipped test does not add useful flakiness signal, so this PR stops scheduling EFD retries when the first execution is skipped && !failed.

Testing

• go test ./ddtrace/mocktracer -count=1
• go test ./ddtrace/mocktracer -race -count=1
• go test ./ddtrace/mocktracer -race -count=3
• go test ./ddtrace/mocktracer -run 'TestCIVisibilityMockTracer' -count=10
• go test ./ddtrace/tracer -run 'TestSetGlobalTracerPreservingCIVisibilityMockTracer|TestGlobalTracer' -count=1
• go test ./ddtrace/tracer -run 'TestSetGlobalTracerPreservingCIVisibilityMockTracer|TestGlobalTracer' -race -count=10
• go test ./ddtrace/tracer -run 'TestCIVisibilityImplementsTraceWriter|TestCiVisibilityTraceWriterFlushRetries|TestCiVisibilityTraceWriterClosesHTTPResponseBody' -count=1
• go test ./ddtrace/tracer -run 'TestCIVisibilityImplementsTraceWriter|TestCiVisibilityTraceWriterFlushRetries|TestCiVisibilityTraceWriterClosesHTTPResponseBody' -race -count=1
• go test ./ddtrace/tracer -run 'TestCIVisibility|TestCiVisibility' -count=1
• go test ./internal/civisibility/integrations/gotesting -count=1
• go test ./internal/civisibility/integrations/gotesting/subtests -count=1
• go test ./internal/civisibility/integrations/gotesting/failnow -count=1
• git diff --check

Local note: full go test ./ddtrace/tracer -count=1 is blocked in this checkout by environment-sensitive DogStatsD autodetection expectations. The local agent reports DogStatsD port 8135, while existing tracer tests expect 8125 in startup-log/default-config assertions. The focused tracer tests for these changes pass.

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.