Support zero-shot CoT for filter

[sidjha1]

Addresses #194. I think the CoT code has become very convoluted and there is probably a simpler way to achieve the desired functionality.

import lotus
from lotus.models import LM
from lotus.types import ReasoningStrategy
import pandas as pd

lm = LM(model="gpt-4.1")
lotus.settings.configure(lm=lm)

data = {
    "Text": [
        "I had two apples, then I gave away one",
        "My friend gave me an apple",
        "I gave away both of my apples",
        "I gave away my apple, then a friend gave me his apple, then I threw my apple away",
    ]
}
df = pd.DataFrame(data)
user_instruction = "{Text} I have at least one apple"
filtered_df = df.sem_filter(
    user_instruction, strategy=ReasoningStrategy.ZS_COT, return_all=True, return_explanations=True
) 

print(filtered_df)

Filtering: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████ 4/4 LM calls [00:03<00:00,  1.17it/s]
                                                Text  filter_label                                 explanation_filter
0             I had two apples, then I gave away one          True  The text states, "I had two apples, then I gav...
1                         My friend gave me an apple          True  The text states, "My friend gave me an apple."...
2                      I gave away both of my apples         False  The text states, "I gave away both of my apple...
3  I gave away my apple, then a friend gave me hi...         False  Let's break down the sequence of events:\n1. "...

[sidjha1]

Never used

[liana313]

LGTM!

Addresses #194. I think the CoT code has become very convoluted and there is probably a simpler way to achieve the desired functionality.

And I agree, opening #198 to address this. PR #140 is also related.