[tamnd]

Part of #1446.

glob(path) is now a table function. It answers one file VARCHAR row for each file a pattern matches and reads none of them, which the corpus uses in 80 records and which rudb refused before as a table function that does not exist.

It takes a pattern or a list of them. The matches of one pattern come out sorted, and a list keeps the order it was written in, duplicates and all, so glob(['b.csv', '*.csv']) gives b.csv, a.csv, b.csv, as the pin does. ** matches at any depth, hidden files are listed, and a directory is not a file, so a pattern that names only directories gives no rows. A path that matches nothing, an empty string and an empty list give no rows rather than an error.

The refusals are the pin's: a NULL argument, a NULL in a list, a list of something other than strings, a number, and a named parameter, each in the pin's words. duckdb_functions() lists the three overloads the pin lists, for VARCHAR, ANY[] and VARIANT.

The paths are worked out in the binder the way read_text and read_blob work them out, so the executor is handed names rather than a pattern. A test in crates/rudb/tests/read_text.rs covers the patterns, the list order, the empty cases, the overloads and the refusals, checked on v2.0.0-dev84237.