fix(aws_s3 source): ignore CloudTrail log delivery notifications

[pupapaik]

Summary

• Customers often point a trail's own SNS topic at the same topic their bucket publishes S3 events to, so our queue gets a second message per log file, shaped {"s3Bucket": ..., "s3ObjectKey": [...]}. It matched no SqsEvent variant, so it was logged as Failed to process SQS message and never deleted; it came back every visibility timeout until retention expired. On gainsight that is ~20-37 errors/s and ~400 undeletable messages/hour, heading for a ~39k in-flight plateau.
• Recognise that shape and treat it like s3:TestEvent: warn, count it under sqs_s3_event_record_ignored_total{ignore_type="cloudtrail_notification"}, and delete it. No customer action needed, and no data impact: the S3 event for the same file is what gets ingested, which is also why the files are deliberately not fetched from this message (that would double every event).
• The new variant is last in the untagged enum, so every body that parsed before resolves to the same variant as before. Bodies that match nothing still fail loudly.

Tests: cargo test --no-default-features --features sources-aws_s3 --lib -- sources::aws_s3::sqs on rust 1.92 -- 7 passed, 0 failed. Three are new: raw shape, SNS-wrapped shape, and a variant-order regression that also asserts near-miss bodies still fail to parse.

Details

A customer with only the trail topic (no S3 event notifications) moves from erroring to silent. That case stays detectable: the Grafana "no events and no errors" rules for source_(cloudtrail|gcp).* (20 min) fire instead of VectorSourceAllErrors, and the new warning plus counter say why.

Rollout after merge: build a new vector-base tag from this branch, then bump the FROM .../vector-base: line in tools/chronoforge/Dockerfile and goservices/vector_exec_sources/Dockerfile in operations. v0.54.0-exaforce has no other undeployed commits, so the bump ships only this change.

Verification on gainsight (both message kinds arrive in eu-prod-exaforce-cloudtrail-logs-queue): SQS deleted/sent should go from 0.51 to ~1.0, ApproximateNumberOfMessagesNotVisible should drain within one visibility timeout, and hourly CloudTrail event counts should not change.

[pupapaik]

@sundaresanr can you check thsi one if it make sense. I see it in gainsight

[sundaresanr]

#48