Fix memory dedup to use exact entry matching

[ZoranPandovski]

The duplicate check in hippocampus used if text in content which matches substrings across the whole file. This causes two problems:
e.g

• "Use httpx" blocks "Use httpx with timeout=15" (false reject)
• "Use httpx" in a rule's metadata or another entry's text blocks unrelated entries (false reject)

Now we parse each - entry line, strip the metadata, and compare exact entry text.

@torrmal I think there is another bug when we do compaction and we loos metadata. I will open separate PR