fix: linear reranker applies wrong score to combine (#2035)

related to #2014 
this fixes:
- linear reranker may lost some results if the merging consumes all
vector results earlier than fts results
- linear reranker inverts the fts score but only vector distance can be
inverted

---------

Signed-off-by: BubbleCal <bubble-cal@outlook.com>

python/python/tests/test_rerankers.py

@@ -3,6 +3,7 @@ import random
 
 import lancedb
 import numpy as np
+import pyarrow as pa
 import pytest
 from lancedb.conftest import MockTextEmbeddingFunction  # noqa
 from lancedb.embeddings import EmbeddingFunctionRegistry
@@ -281,6 +282,31 @@ def _run_test_hybrid_reranker(reranker, tmp_path, use_tantivy):
 @pytest.mark.parametrize("use_tantivy", [True, False])
 def test_linear_combination(tmp_path, use_tantivy):
     reranker = LinearCombinationReranker()
+
+    vector_results = pa.Table.from_pydict(
+        {
+            "_rowid": [0, 1, 2, 3, 4],
+            "_distance": [0.1, 0.2, 0.3, 0.4, 0.5],
+            "_text": ["a", "b", "c", "d", "e"],
+        }
+    )
+
+    fts_results = pa.Table.from_pydict(
+        {
+            "_rowid": [1, 2, 3, 4, 5],
+            "_score": [0.1, 0.2, 0.3, 0.4, 0.5],
+            "_text": ["b", "c", "d", "e", "f"],
+        }
+    )
+
+    combined_results = reranker.merge_results(vector_results, fts_results, 1.0)
+    assert len(combined_results) == 6
+    assert "_rowid" in combined_results.column_names
+    assert "_text" in combined_results.column_names
+    assert "_distance" not in combined_results.column_names
+    assert "_score" not in combined_results.column_names
+    assert "_relevance_score" in combined_results.column_names
+
     _run_test_hybrid_reranker(reranker, tmp_path, use_tantivy)