test(strings): improve selected string tests

[gforsyth]

BigQuery:

• rstrip and lstrip are broken because of a SQLGlot bug, already fixed upstream.

Snowflake:

• rstrip and lstrip are broken because of a SQLGlot bug, already fixed upstream.

MSSQL:

• contains might work, but requires “fulltext index” and I can’t be bothered
• rpad not defined
• lpad not defined
• find_in_set not defined

Oracle:

• Treats len(🐍) == 2, should be able to use LENGTHC instead
• strip maps to trim, but TRIM doesn’t accept custom characters, unlike RTRIM and LTRIM

Flink:

• Treats len(🐍) == 2
• strip maps to trim, but TRIM doesn’t accept custom characters, unlike RTRIM and LTRIM
  There is a function in the docs called BTRIM but maybe it’s only in dev? But that should work?

Impala (woof):

• Treats len(🐍) == 4, len(Éé) == 4
• No TRIM for anything but whitespace. This appears to be a limitation of our Impala compiler.
• Accented characters do not get upper-cased or lower-cased, just left as-is.

Risingwave

• Accented characters do not get upper-cased or lower-cased, just left as-is.

MySQL

• Treats len(🐍) == 4, should use CHAR_LENGTH

PySpark

• TRIM only trims spaces

SQLite

• Accented characters do not get upper-cased or lower-cased, just left as-is.

Clickhouse

• Lengths of accented characters and emoji are of byte width, not character length
• This also impacts padding
  BUT:
  There are UTF8 aware versionf for all of these

What should the string APIs do?

find_in_set

We state in the docs for find_in_set that if a value contains a comma, we
return -1 (the same behavior as if there is no match). Currently no backend
does this, the 5 backends that support the operation all return the index of the
comma-containing field (duckdb, datafusion, mysql, postgres,
risingwave)

We can just update the docstring?

Padding

The SQL convention with padding is that if the original string is longer than
the pad length, the string is trimmed to that length.

The Python convention is to leave strings >= pad_length alone.

Which would we like to adopt?

I think the SQL convention here is terrible and we should follow Python.

UTF-8

Some backends will support this, some won’t, but, we should be consistent.

I think that length, lpad, and rpad should all be UTF-8 variants.

Propose adding len_bytes as an additional string method for the byte length of a string.

[gforsyth]

Ok, I've fixed strip, lstrip and rstrip for all backends except:

• pyspark: because they don't have upstream rtrim and ltrim functions that take characters to trim
• snowflake: upstream fixed sqlglot bug, just waiting on it to release
• bigquery: upstream fixed sqlglot bug, just waiting on it to release

I'm going to open issues to track the other proposed work in this PR description and handle them separately, since some of those might be breaking changes, while everything here is strictly a bugfix.

[cpcloud]

🚢 it!

[cpcloud]

Doing a rebase merge to pick up all the individual fixes.