JITSU-138: fix Datadog reserved-status collision; synthesize otlp message lines

[absorbb]

Follow-up to JITSU-138, from end-to-end testing against Datadog's agentless OTLP intake (which merges the structured body's top-level keys into root log attributes).

Architecture (reworked per review): the otlp destination stays fully payload-agnostic — otlp.go is untouched relative to newjitsu. Body adaptation happens where envelopes are produced to the Kafka topic, and only for bodies the pipeline constructs itself:

1. Go producer (eventslog/kafka_events_log.go), for bulker_batch / bulker_stream records (bulker-built bulker.State / stream-status shapes):
   • top-level status → record_status: COMPLETED/FAILED values collide with Datadog's reserved status attribute and made every record render as critical regardless of severity (reproduced in isolation with controlled probe payloads)
   • synthesized message when absent/empty/null: bulker_batch COMPLETED: 2 rows → events39, with a rune-safe-truncated error preview on failures
   • the eventId hash is computed from the unadapted body — ids and billing dedup unchanged
2. Rotor producer (kafka-events-store.ts): dead-letter bodies (rotor-built {payload, error}) get their message line at construction.

Function-log bodies are user data and are never touched — no key renames, no synthesized fields.

Body stays structured (kvlist) — no stringification. Covered by TestKafkaEventsLogAdaptsOwnedBodies (rename, synthesis, truncation, function-body-untouched) and rotor deadLetterMessage tests (incl. surrogate-pair safety).

Docs note: jitsucom/websites#60.

🤖 Generated with Claude Code

[jitsu-code-review]

Reviewed the OTLP body-shaping changes and the new coverage in otlp_test.go (including the status remap, synthetic message generation, and copy semantics). I found two potential regressions worth a quick follow-up in inline comments.

[jitsu-code-review]

Reviewed the OTLP body-presentation changes in otlp.go and the new coverage in otlp_test.go. I left inline comments for a few correctness/regression risks worth clarifying: unconditional status remapping on generic OTLP payloads, message: null bypassing synthesis, and potential overwrite when record_status already exists.

[jitsu-code-review]

Reviewed the producer-side adaptation changes in and , plus their new tests. I focused on correctness, regressions, and user-visible behavior (status remapping/message synthesis scope and eventId/billing implications). I didn’t find actionable bugs or security issues in this diff.

[jitsu-code-review]

Reviewed producer-side adaptation changes in bulker/eventslog/kafka_events_log.go and services/rotor/src/lib/kafka-events-store.ts, plus the added tests. I focused on correctness, regressions, and user-visible behavior (status remapping/message synthesis scope and eventId/billing implications). I did not find actionable bugs or security issues in this diff.

[jitsu-code-review]

Reviewed the changes in bulker/eventslog and services/rotor for correctness, regressions, and security impact.

I focused on the producer-side body adaptation (status→record_status, synthesized message) and dead-letter message synthesis/truncation behavior. I did not find actionable bugs or user-visible regressions in this patch.

Sanity check: go test ./eventslog passes in bulker/.