fix: ManualKeep should override the propagated sampling decision (?)

[darccio]

What does this PR do?

Motivation

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 ./scripts/lint.sh locally.
• Add an appropriate team label so this PR gets put in the right place for the release notes.
• Non-trivial go.mod changes, e.g. adding new modules, are reviewed by @DataDog/dd-trace-go-guild.

Unsure? Have a question? Request a review!

[datadog-official]

⚠️ Tests

⚠️ Warnings

🧪 2 Tests failed

TestRulesSampler from github.com/DataDog/dd-trace-go/v2/ddtrace/tracer (Datadog)

Failed

=== RUN   TestRulesSampler
--- FAIL: TestRulesSampler (0.24s)

TestRulesSampler/manual_keep_on_propagated_sampling_decision from github.com/DataDog/dd-trace-go/v2/ddtrace/tracer (Datadog)

Failed

=== RUN   TestRulesSampler/manual_keep_on_propagated_sampling_decision
    sampler_test.go:1179: 
        	Error Trace:	/home/runner/work/dd-trace-go/dd-trace-go/ddtrace/tracer/sampler_test.go:1179
        	Error:      	Not equal: 
        	            	expected: int(2)
        	            	actual  : float64(-1)
        	Test:       	TestRulesSampler/manual_keep_on_propagated_sampling_decision
    --- FAIL: TestRulesSampler/manual_keep_on_propagated_sampling_decision (0.01s)

ℹ️ Info

❄️ No new flaky tests detected

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 00bed5a | Docs | Was this helpful? Give us feedback!}

[pr-commenter]

Benchmarks

Benchmark execution time: 2025-10-23 16:49:33

Comparing candidate commit 00bed5a in PR branch push-lmzoulmktnut with baseline commit 1eb58b4 in branch main.

Found 0 performance improvements and 0 performance regressions! Performance is the same for 24 metrics, 0 unstable metrics.