fix: enforce direct JIT decorator in optimizer prompt for numerical code

When is_numerical_code is true, the LLM sometimes outputs conditional
fallback paths (try/except, if/else) instead of applying the JIT
decorator directly. Add explicit output format instructions to prevent
this behavior.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

django/aiservice/core/languages/python/optimizer/jit_instructions.md

@@ -47,6 +47,15 @@ Many Python libraries are not JIT-compatible:
 ### 7. Functions with Heavy Object Creation
 Code that creates many Python objects, uses class instances extensively, or relies on Python's dynamic nature will not JIT well.
 
+## JIT Output Format (CRITICAL)
+
+When you determine that JIT compilation is viable for the given code:
+- Apply the JIT decorator **directly** to the output function (e.g., `@numba.njit`, `@torch.compile`, `@tf.function`, `@jax.jit`).
+- Add the necessary import (e.g., `import numba`) at the top of the code.
+- Do **NOT** create conditional fallback paths. Never wrap JIT usage in `try/except ImportError`, `if HAS_NUMBA`, or any similar if/else branching that falls back to a non-JIT version.
+- Do **NOT** create a separate "fast" helper function alongside the original. The output must be a single, clean function with the JIT decorator applied.
+- If JIT is not viable for this code, optimize it using other strategies without any JIT decorators — do not include a JIT path at all.
+
 ## Guidelines for Numba (`@njit`)
 Use Numba when the code:
 - Performs numerical computations with NumPy arrays