pretext: vectorize field on pipeline + PUT /pipelines/{id}/vectorize

[beckyconning]

Summary

vectorization-spec-v4 / LD13 + LD15 API surface:

• Pipeline schema gains vectorize: boolean (nullable, optional). When true, loads of datasets in the pipeline produce 384-dim embeddings landed in a sibling PRETEXT table on the destination. Absent or false preserves pre-feature behaviour byte-identically.
• New endpoint PUT /pipelines/{id}/vectorize for runtime opt-in/out at pipeline scope (mirrors the pii-config pattern).
• POST /pipelines and the new PUT both return 400 when vectorize=true is set against a destination that does not support the Snowflake VECTOR column type (LD15 — Snowflake-family destinations only).

Companion PRs

• precog/services#3478 — implementation
• precog/precog-destination-snowflake#493 — public SnowflakeKind.parseConfigValues
• precog/devops#1536 — staging pin

Verification

End-to-end verified on staging via new pipeline customer service vectorize (Intercom Conversations source, vectorize: true):

SELECT COUNT(*) FROM procurement_staging.CUSTOMER_SERVICE_VECTORIZE.PRETEXT;
-- 180

SELECT VECTOR_COSINE_SIMILARITY(vec, vec) FROM PRETEXT LIMIT 1;
-- ~1.0

Merge-readiness summary: Linear PRD-212.

Test plan

• Confirm OpenAPI spec validates clean.
• Confirm the new vectorize field is correctly typed as nullable boolean.
• Confirm PUT /pipelines/{id}/vectorize request/response shapes match what the implementation in precog/services#3478 ships.

🤖 Generated with Claude Code