feat(strings): `len_bytes` and `length` for string columns

[gforsyth]

As turned up in #10020, we get variable answers back from backends regarding the length of strings, once those strings aren't just simple ASCII.

This is due to the length of a string being sort of a fuzzy concept in the age of Unicode.

I think we should try to standardize on:

• length returns the count of the number of characters in the string, so e.g. len(🐍) == 1 and len(É) == 1

This may not be possible on all backends, but I think we can get close, then add a new method

• len_bytes (I'm open to naming suggestions) that returns the length in bytes of the string, len_bytes(🐍) == 4 and len(É) == 2

An incomplete list of functions we can use on various backends for character length:

• Oracle: LENGTHC
• MySQL: CHAR_LENGTH
• Clickhouse: lengthUTF8
• Impala: CHAR_LENGTH (maybe...)

[cpcloud]

Here's a table of how our backends handle computing the length of the string 🤦‍♂️🤦‍♀️. It will surprise no one that not every backend we support agrees on this string's length.

TL; DR: consensus seems to be to count the number of code points

The code points are:

>>> s = '🤦‍♂️🤦‍♀️'
>>> code_points = [c.encode("unicode_escape") for c in s]
>>> print('\n'.join(map(bytes.decode, code_points)))
\U0001f926  # person facepalming
\u200d      # zero-width joining character
\u2642      # male symbol
\ufe0f      # variation selector 16
\U0001f926  # person facepalming
\u200d      # zero-width joining character
\u2640      # female symbol
\ufe0f      # variation selector 16
>>> len(code_points)
8

backend	function	len('🤦‍♂️🤦‍♀️')	notes
bigquery	CHAR_LENGTH	8
clickhouse	lengthUTF8	8
datafusion	LENGTH	8
druid	LENGTH	10	probably related to UTF-16
duckdb	LENGTH	8
exasol	LENGTH	8
flink	CHAR_LENGTH	8
impala	CHAR_LENGTH	26	this is the byte length
mssql	LEN	6	no idea
mysql	CHAR_LENGTH	14	byte length of non zero width chars?
oracle	LENGTHC	6	no idea
polars	col.str.len_chars()	8
postgres	CHAR_LENGTH	8
pyspark	length	8
risingwave	CHAR_LENGTH	8
snowflake	LENGTH	8
sqlite	LENGTH	8
trino	LENGTH	8

[cpcloud]

The upshot here is that we can't make any strong guarantees about what is returned across backends for string operations that depend on a notion of "character", even when encoding is held constant.

This is why we can't have nice strings.

[gforsyth]

:chefs_kiss:

[gforsyth]

Yeah, no strong guarantees, but we can get the existing ones closer to each other, and then have len_bytes available if someone is specifically looking for that

[cpcloud]

Yep, sounds good to me. Slight preference for byte_length over len_bytes or length_bytes, OR deprecating length in favor of len and adding len_bytes or byte_len.

[cpcloud]

Actually, I hate byte_len, so scratch that one from my suggestion list.