[BUG] loc-based indexing of DataFrames silently discards missing keys if at least one key is present in indexer

[wence-]

Describe the bug

When performing loc indexing of a DataFrame, if one asks for a missing key, pandas raises a KeyError. In constrast, cudf only does so if none of the requested keys are in the index. If at least one requested key is present then the subsetted data frame with that key is returned and the missing keys are silently dropped.

Series loc indexing does the right thing here and raises KeyError if any keys are missing.

Steps/Code to reproduce bug

import pandas as pd
import cudf

# same failure with rangeindex too.
df = pd.DataFrame({"A": range(5)}, index=list(range(5)))
cdf = cudf.from_pandas(df)

df.loc[[0, 5]] # 5 is missing, raises KeyError

cdf.loc[[0, 5]]
#    A
# 0  0

Expected behavior

Should match pandas behaviour and raise KeyError.

[wence-]

TBF, pandas used to behave like this too (but it's been removed for a while).