rows_from_file() splits a single-column CSV file on a letter, or raises csv.Error #907

When rows_from_file() detects the format of a single-column CSV file, it either fails or splits the column on a letter. That affects sqlite-utils memory file.csv and any Python call that leaves format= unset.

$ printf 'name\nAnn\nBob\n' > names.csv
$ sqlite-utils memory names.csv "select * from names"
...
_csv.Error: Could not determine delimiter

$ printf 'url\nhttps://example.com/a\nhttps://example.org/b\n' > urls.csv
$ sqlite-utils memory urls.csv "select * from urls"
[{"ur": "https://examp", "": "e.com/a"},
 {"ur": "https://examp", "": "e.org/b"}]

(sqlite-utils 4.2.1)

The detection calls csv.Sniffer().sniff() on the first 2048 bytes without a delimiters= list. The sniffer then treats any character as a candidate, so a file with a single column either has no consistent candidate (csv.Error) or has a letter that happens to repeat the same number of times on every line (l above).

A fix could sniff only for the delimiters a CSV file is realistically written with (,, ;, tab, |) and read the file as one column when none of them is found. insert --sniff calls the sniffer the same way, but that is an explicit request to detect the dialect, so I would leave it unchanged.