feat(qdrant): index content, is_active, and chunk_index on the chunks collection

Why:
- ADR-0002's keyword search needs a full-text index on content, which
  collection.py deliberately deferred to plan 002. is_active and chunk_index
  were unindexed while ingestion was the only reader; every /v1/points read path
  filters on them.

Changes:
- content gets a TEXT index with the multilingual tokenizer, which segments
  Persian correctly where the word tokenizer mishandles ZWNJ-joined compounds.
  No stemmer or stopword list: content is already letter-folded by
  normalize_persian_text at ingest, and the ranked Farsi lexical path is the
  benchmarked BM25 sparse vector, not this index.
- Tests assert content is TEXT rather than KEYWORD -- a keyword index would only
  match an entire chunk verbatim, which never happens and fails silently.
- Adds a test that a missing index is added to an already-live collection.

Impact:
- Requires re-running `python -m src.cli.qdrant_bootstrap`. Payload indexes are
  additive, so no collection rebuild and no re-embedding.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

src/infrastructure/qdrant/collection.py

@@ -80,14 +80,25 @@ def _sparse_vectors_config() -> dict[str, models.SparseVectorParams]:
     return {SPARSE_VECTOR: models.SparseVectorParams(modifier=models.Modifier.IDF)}
 
 
+# `content`'s full-text index backs ADR-0002's keyword search. The
+# `multilingual` tokenizer is the one that segments Persian correctly; `word`
+# splits on non-alphanumerics, which mis-handles ZWNJ-joined compounds. No
+# stemmer or stopword list is configured: `content` is already letter-folded by
+# `normalize_persian_text` at ingest (ADR-0018), and the *ranked* Farsi lexical
+# path is the benchmarked BM25 sparse vector, not this index. This one exists
+# for exact keyword/filter matching, which ADR-0002 keeps deliberately distinct
+# from retrieval.
+_CONTENT_INDEX = models.TextIndexParams(
+    type=models.TextIndexType.TEXT,
+    tokenizer=models.TokenizerType.MULTILINGUAL,
+    lowercase=True,
+)
+
 # (field name, schema). `order_id` is float because Qdrant's `Range` conditions
 # and `order_by` only support numeric/datetime payloads -- a keyword key could
 # only be sorted client-side after fetching every chunk (ADR-0001).
-#
-# The full-text index on `content` is deliberately absent: it belongs to plan
-# 002's keyword search, and payload indexes -- unlike vector config -- can be
-# added to a live collection later.
-_PAYLOAD_INDEXES: tuple[tuple[str, models.PayloadSchemaType | models.KeywordIndexParams], ...] = (
+_PayloadIndexSchema = models.PayloadSchemaType | models.KeywordIndexParams | models.TextIndexParams
+_PAYLOAD_INDEXES: tuple[tuple[str, _PayloadIndexSchema], ...] = (
     # `is_tenant` co-locates a tenant's vectors on disk for sequential reads,
     # which is the whole point of payload-partitioned multitenancy.
     (
@@ -99,6 +110,13 @@ _PAYLOAD_INDEXES: tuple[tuple[str, models.PayloadSchemaType | models.KeywordInde
     ("order_id", models.PayloadSchemaType.FLOAT),
     ("previous_chunk_id", models.PayloadSchemaType.KEYWORD),
     ("next_chunk_id", models.PayloadSchemaType.KEYWORD),
+    ("content", _CONTENT_INDEX),
+    # Every read path filters on `is_active` (ADR-0002 implies `is_active: true`
+    # unless the caller opts in), and the re-ingestion sweep and `/v1/points`
+    # both range over `chunk_index`. Both were unindexed while ingestion was the
+    # only reader; plan 002 makes them hot.
+    ("is_active", models.PayloadSchemaType.BOOL),
+    ("chunk_index", models.PayloadSchemaType.INTEGER),
 )
 
 

tests/integration/qdrant/test_collection.py

@@ -65,10 +65,42 @@ async def test_ensure_chunks_collection_creates_the_payload_indexes(
         "order_id",
         "previous_chunk_id",
         "next_chunk_id",
+        "content",
+        "is_active",
+        "chunk_index",
     }
     # order_id must be numeric: Qdrant's Range/order_by reject keyword payloads.
     assert schema["order_id"].data_type == models.PayloadSchemaType.FLOAT
     assert schema["tenant_id"].data_type == models.PayloadSchemaType.KEYWORD
+    # `content` must be TEXT, not KEYWORD: ADR-0002's keyword search is a
+    # full-text match on it, and a keyword index would only match the entire
+    # chunk verbatim -- which never happens and would fail silently.
+    assert schema["content"].data_type == models.PayloadSchemaType.TEXT
+    assert schema["is_active"].data_type == models.PayloadSchemaType.BOOL
+    assert schema["chunk_index"].data_type == models.PayloadSchemaType.INTEGER
+
+
+async def test_ensure_chunks_collection_adds_a_missing_index_to_a_live_collection(
+    qdrant_client: AsyncQdrantClient, qdrant_settings: QdrantSettings
+) -> None:
+    """Payload indexes are additive, unlike vector config.
+
+    This is the operational claim the runbook makes when a new index ships:
+    re-running the bootstrap against an existing collection adds it in place, so
+    plan 002's `content`/`is_active`/`chunk_index` indexes do not require
+    recreating a collection that already holds a tenant's points.
+    """
+    await ensure_chunks_collection(qdrant_client, collection=qdrant_settings.collection)
+    await qdrant_client.delete_payload_index(
+        collection_name=qdrant_settings.collection, field_name="content"
+    )
+    info = await qdrant_client.get_collection(qdrant_settings.collection)
+    assert "content" not in info.payload_schema
+
+    assert not await ensure_chunks_collection(qdrant_client, collection=qdrant_settings.collection)
+
+    info = await qdrant_client.get_collection(qdrant_settings.collection)
+    assert info.payload_schema["content"].data_type == models.PayloadSchemaType.TEXT
 
 
 async def test_ensure_chunks_collection_is_idempotent(