feat(eap-items): ingest conversation_id and session_id

[phacops]

Summary

Populates the conversation_id and session_id columns in the eap-items processor from the new TraceItem proto fields.

The columns themselves already exist — they were added as non-nullable UUID columns (after trace_id) by migration 0060_add_conversation_id_and_session_id (#8053) and are declared in the eap_items storage schema — but nothing was writing to them. This change wires the ingestion path so the values actually land in ClickHouse when present on the TraceItem.

Changes

• Upgrade sentry-protos to 0.34.0 for both Python (pyproject.toml + uv.lock) and Rust (Cargo.toml + Cargo.lock). 0.33.0 is the first version that exposes conversation_id (field 10) and session_id (field 11) on TraceItem; 0.34.0 is the current latest and contains no breaking changes vs 0.32.0.
• Ingest the new fields in rust_snuba/src/processors/eap_items.rs:
  • Add conversation_id / session_id to both EAPItem and the RowBinary EAPItemRow (with the existing ClickHouse UUID serde adapter), inserted right after trace_id to match the on-disk column order.
  • Add both to EAPItemRow::COLUMN_NAMES so the explicit INSERT ... FORMAT RowBinary column list stays aligned with the struct's wire order.
  • The proto fields arrive as strings. A parse_uuid_or_random helper parses them into UUIDs.

Behavior — absent ids are randomized, not zero

The columns are non-nullable UUID. Rather than persisting the all-zero "magic" UUID when a TraceItem has no conversation_id/session_id:

• Present and valid (non-nil) → stored as the parsed UUID.
• Absent (empty string), malformed, or explicitly all-zero → replaced with a fresh random UUID.

The all-zero value would be indistinguishable from a real (if absurd) all-zero id and would balloon the bf_conversation_id/bf_session_id bloom-filter indexes with a single high-frequency value. Randomizing keeps the columns non-nullable, so no schema migration is required. (Nullable(UUID) was considered but rejected to avoid altering the just-merged migration 0060.)

Tests

• Added unit tests covering present / absent / malformed / explicit-nil values on both the typed conversion and the RowBinary path.
• Updated test_column_names_match_struct_layout for the two new columns and shifted indices.
• Updated the test_schemas insta snapshot and added redactions for the two now-random fields so it stays deterministic.
• cargo build passes and all processors unit tests pass. The one exception is test_row_binary_clickhouse_insert, which requires a live ClickHouse instance (not available in the dev sandbox); it runs in CI and exercises the full RowBinary insert against the real table — including the new column ordering.

🤖 Generated with Claude Code