[FEAT] Add string tokenize expression

[Vince7778]

Allows users to tokenize a string column using tiktoken and a variety of encoders.

Todo list:

• Support for builtin models (cl100k_base, p50k_base, etc)
• Support for loading models from a token file
• Support for downloading models from the cloud
• More tests
• Fix error handling
• Pattern argument
• Special token support
• All the tests
• Update docs

Things that could be done in the future:

• Add caching for token files so that the processes don't have to each download it.
• Make a fork of tiktoken-rs for various fixes (accessing private fields, fixing unwraps, optimizing etc)
• Add support for huggingface tokenizers
• Add more granular special token support (custom inputs, using a subset of them)

[codecov]

Codecov Report

Attention: Patch coverage is 90.71926% with 40 lines in your changes missing coverage. Please review.

Please upload report for BASE (main@346868e). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2503   +/-   ##
=======================================
  Coverage        ?   63.31%           
=======================================
  Files           ?      978           
  Lines           ?   108843           
  Branches        ?        0           
=======================================
  Hits            ?    68915           
  Misses          ?    39928           
  Partials        ?        0           

Files	Coverage Δ
daft/expressions/expressions.py	93.82% <100.00%> (ø)
src/daft-core/src/series/ops/utf8.rs	93.54% <100.00%> (ø)
src/daft-functions/src/lib.rs	53.33% <100.00%> (ø)
src/daft-functions/src/tokenize/mod.rs	100.00% <100.00%> (ø)
src/daft-functions/src/tokenize/special_tokens.rs	100.00% <100.00%> (ø)
src/daft-functions/src/tokenize/bpe.rs	96.35% <96.35%> (ø)
src/daft-functions/src/tokenize/encode.rs	87.09% <87.09%> (ø)
src/daft-functions/src/tokenize/decode.rs	73.56% <73.56%> (ø)

[universalmind303]

@Vince7778 Do we have any plans on eventually supporting huggingface tokenizers as well?

[Vince7778]

I think for now I won't, since huggingface has a much different API and a lot more customization. Seems like it would be a pretty big chore to support it. If there's enough demand we probably can but for now using UDFs would be better.