feat: Ported back new grammar changes from C++ to Python implementation (#1637)

* Backported . (any chat) from llama.cpp

* unfinished {count,optionalmax)

* implemented slice function in std:vector

* fixed mistake done while reading

* ported ggml-org/llama.cpp#7194

* multiple fixes, var copy

* Rewrite LlamaGrammar internals in python style

* bugfix

---------

Co-authored-by: Andrei <[email protected]>

Author: NoBrainer24

llama_cpp/llama_cpp.py

@@ -3002,7 +3002,7 @@ def llama_grammar_init(
     n_rules: Union[ctypes.c_size_t, int],
     start_rule_index: Union[ctypes.c_size_t, int],
     /,
-) -> llama_grammar_p:
+) -> Optional[llama_grammar_p]:
     """Initialize a grammar from a set of rules."""
     ...