Clean up the __slots__ and overall inheritance structure in _grammar.py

[hudson-ai]

This PR primarily serves to align the attributes of our grammar classes with the arguments they are able to pass to the llguidance api.

The most notable change is that max_tokens is only allowed on classes that will serialize to Gen, Lexeme, or GenGrammar, where token_limit is reimplemented by wrapping a grammar in a GenGrammar pre-image (Subgrammar).

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

Attention: Patch coverage is 91.66667% with 5 lines in your changes missing coverage. Please review.

Project coverage is 65.34%. Comparing base (c92d80a) to head (f7c385b).

Files with missing lines	Patch %	Lines
guidance/_grammar.py	91.37%	5 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1089      +/-   ##
==========================================
- Coverage   66.67%   65.34%   -1.33%     
==========================================
  Files          65       65              
  Lines        5173     5151      -22     
==========================================
- Hits         3449     3366      -83     
- Misses       1724     1785      +61     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[hudson-ai]

Closing PR, as I'm going to combine it with the PR switching to llguidance JSON