[CI] Revert back prepare_prompts and check_answers

[WoosukKwon]

No description provided.

[gemini-code-assist]

Code Review

This pull request reverts a previous change, moving the prep_prompts and check_answers test utility functions back into tests/utils.py.

I've reviewed the reverted code and found two high-severity issues:

1. In prep_prompts, the use of random.seed() affects the global random state, which can lead to flaky tests. I've suggested using an isolated random.Random instance to ensure deterministic behavior without side effects.
2. In check_answers, the logic for parsing model outputs is brittle and can fail with leading whitespace. I've proposed a more robust parsing method.

These changes will improve the reliability and maintainability of the test suite.

[gemini-code-assist]

Using random.seed() has a global effect and can interfere with other tests that rely on the random number generator. It's better to use an isolated random.Random instance for deterministic generation within this function.

Additionally, the prompt string construction can be simplified into a single f-string for better readability.

Suggested change

	random.seed(1)
	for _ in range(batch_size):
	idx = random.randint(30, 90)
	indices.append(idx)
	prompt = "```python\n# We set a number of variables, " + \
	f"x{idx} will be important later\n"
	ln = random.randint(*ln_range)
	for k in range(30, ln):
	v = random.randint(10, 99)
	rng = random.Random(1)
	for _ in range(batch_size):
	idx = rng.randint(30, 90)
	indices.append(idx)
	prompt = f"```python\n# We set a number of variables, x{idx} will be important later\n"
	ln = rng.randint(*ln_range)
	for k in range(30, ln):
	v = rng.randint(10, 99)

[gemini-code-assist]

The current logic for parsing the model's output is brittle. If the output has leading whitespace, for example " 55" instead of "55", text[0:2] would be " 5", which incorrectly parses to 5.

A more robust approach is to strip whitespace from the whole string first, then split it and take the first part. This correctly handles leading/trailing whitespace.

Note that this can still raise an IndexError for empty outputs or a ValueError if the output is not a number, but this might be an acceptable risk for this test's context.

Suggested change

	answer2 = [int(text[0:2].strip()) for text in outputs]
	answer2 = [int(text.strip().split()[0]) for text in outputs]