reject colwise csv map rows with more cells than columns

[uwezkhan]

The colwise reader for std::map<std::string, std::vector> (from<CSV, readable_map_t>::op in include/glaze/csv/read.hpp) advances its per-row loop only past a comma, CR, or LF once it has read n_keys cells, and never looks at ctx.error between fields. A data row carrying more cells than there are header columns, or a field the reader stopped on a stray byte, leaves the iterator on a byte neither the delimiter nor the newline branch consumes, so the outer while(it != end) loop makes no progress.

Before, a,b\n1,2,3\n into a string-valued map read as success with the third cell misassigned to a second row of column a and an empty inserted into b, and the same row of a numeric map set parse_number_failure that the loop dropped and then spun on, emplacing default elements. After, both are a syntax_error. The fix bails on ctx.error after each field and rejects a row not closed by a newline or end of input, which is what the sibling colwise struct reader and the rowwise readers already do; keeping the check in this loop is why the four readers now agree on what a row is. The tradeoff is that a row with trailing content past its columns is now refused rather than partially consumed; a well-formed document, including a last row with no trailing newline, is unchanged.

[stephenberry]

Thanks for the fix! I pushed a small cleanup: the row ending now uses csv_new_line, matching the struct colwise reader, and the numeric extra-cell test checks for syntax_error.