Is there a corpus of toy datasets specifically designed for finding bugs in data science software?

I'm looking for a corpus of toy tabular datasets that can be used to test data profiling, machine learning, data manipulation, etc. software. Some example attributes:

• Strange column names (empty string, long names, duplicate names, names with spaces, periods, syntax, escaped delimiters and tokens)
• Non-rectangular
• Mixed scientific notation in floats, inf literals
• Row-empty or column-empty
• Mixed file encodings
• Numeric and string values designed to overflow memory buffers/cause truncation/rounding to int
• Ambiguous and invalid dates
• Diacritics, emojis

I was going to build a corpus myself, but surely there is some prior work here?