perf: use Haiku model for testgen repair (#2471)

## Summary
- Switch testgen repair endpoint from `EXECUTE_MODEL` (GPT-5-Mini) to
`HAIKU_MODEL` (Haiku 4.5)
- Matches the review endpoint which already uses Haiku
- Repair is a structured task (splice functions, fix assertions) that
doesn't need a frontier model
- Should reduce latency (was timing out at 90s in CI) and cost

django/aiservice/core/shared/testgen_review/repair.py

@@ -25,7 +25,7 @@ from openai.types.chat import (
 
 from aiservice.analytics.posthog import ph
 from aiservice.common.markdown_utils import extract_code_block_with_context
-from aiservice.llm import EXECUTE_MODEL, calculate_llm_cost, call_llm
+from aiservice.llm import HAIKU_MODEL, calculate_llm_cost, call_llm
 from authapp.auth import AuthenticatedRequest
 from core.shared.testgen_review.models import TestRepairErrorSchema, TestRepairResponseSchema, TestRepairSchema
 
@@ -65,7 +65,7 @@ async def testgen_repair(
             obs_context["call_sequence"] = data.call_sequence
 
         response = await call_llm(
-            llm=EXECUTE_MODEL,
+            llm=HAIKU_MODEL,
             messages=messages,
             call_type="testgen_repair",
             trace_id=data.trace_id,
@@ -73,7 +73,7 @@ async def testgen_repair(
             context=obs_context,
         )
 
-        cost = calculate_llm_cost(response.raw_response, EXECUTE_MODEL)
+        cost = calculate_llm_cost(response.raw_response, HAIKU_MODEL)
         logging.debug(f"testgen_repair LLM cost: {cost}")
 
         repair_text = response.content.strip()