feat(evidence): use server-side Qdrant BM25 retrieval

harness/tht/cli/preprocess_cmd.py

@@ -14,6 +14,14 @@ from tht.cli.config_cmd import CONFIG_OPT
 preprocess_app = typer.Typer(help="Materialize versioned preprocessing artifacts")
 
 
+def _bm25_language(workspace_language: str) -> str:
+    languages = {"en": "english", "it": "italian"}
+    try:
+        return languages[workspace_language]
+    except KeyError as exc:
+        raise ValueError("workspace language is unsupported for Qdrant BM25") from exc
+
+
 def _evidence_json_context(config: Path):
     from tht.cli.schema_cmd import _load_config_or_exit
 
@@ -112,6 +120,7 @@ def run_from_config(config: Path, *, dry_run: bool = False, resume: str | None =
         chunk_policy=ChunkPolicy(version="chunk-v1", max_chars=cfg.vector.max_chunk_chars),
         pipeline_version="evidence-v1",
         retain_published_generations=cfg.vector.retain_published_generations,
+        sparse_language=_bm25_language(cfg.language),
     )
     def fingerprint(value: str) -> str:
         return "sha256:" + hashlib.sha256(value.encode()).hexdigest()
@@ -145,6 +154,7 @@ def gc_from_config(config: Path, *, dry_run: bool = False):
         chunk_policy=ChunkPolicy(version="chunk-v1", max_chars=cfg.vector.max_chunk_chars),
         pipeline_version="evidence-v1",
         retain_published_generations=cfg.vector.retain_published_generations,
+        sparse_language=_bm25_language(cfg.language),
     )
     pipeline.workspace_id = cfg._workspace_id
     return pipeline.gc(workspace_root=corpus_root.parent, dry_run=dry_run)