fix(FH-009): include downstream target in memory poisoning path

[Ayorinha]

Summary

• Complete the reported attack path for FH-009 memory-poisoning findings.
• Extend the finding path from the agent reading poisoned shared memory to the concrete downstream target.
• Add regression coverage to ensure the external sink remains visible in the reported path.

Problem

FH-009 correctly detected an untrusted write into shared memory followed by a second agent reading that memory.

However, the reported path could stop at the reading agent even when a downstream target was already identified by the analysis.

This made the attack-path evidence incomplete and could obscure the final sink when reviewing a finding.

Fix

The rule now resolves the downstream path from the reading agent to the identified target and includes that path in the finding.

This preserves the existing detection logic while making the resulting attack-path evidence complete.

Tests

Added regression coverage verifying that:

• FH-009 still fires for the canonical shared-memory poisoning scenario.
• The reported path reaches the downstream external output.
• The final "send" edge is included in the finding.

Scope

Focused change to FH-009 path reporting and its regression test. No changes to unrelated detection rules or runtime behavior.

[Ayorinha]

Thanks for the review opportunity. I’ve kept this PR intentionally focused on the FH-009 attack-path reporting gap and included regression coverage for the downstream sink. The branch is ready for review; I’m happy to adjust anything you’d like changed.