feat: performance improvement by lazy python

[benfdking]

• make comments and texts getters lazy
• adds optimizations to extract_string of when to compute things

[benfdking]

While this makes the tokenizer much quicker, I think it would be good to benchmark slightly higher level to make sure we aren't seeing the slow down in other calls upstream.

[tobymao]

can you run the regular benchmarks?

[benfdking]

Struggling when running python benchmarks/bench.py, getting

'Tokenizer' object has no attribute '_rs_dialect_settings'
'Tokenizer' object has no attribute '_rs_dialect_settings'
'Tokenizer' object has no attribute '_rs_dialect_settings'
'Tokenizer' object has no attribute '_rs_dialect_settings'
Traceback (most recent call last):
  File "/Users/benjaminking/Developer/sqlglot/benchmarks/bench.py", line 226, in <module>
    {k: v if v == "Query" else str(row[k])[0:7] + diff(row, k) for k, v in row.items()}
                                                  ^^^^^^^^^^^^
  File "/Users/benjaminking/Developer/sqlglot/benchmarks/bench.py", line 199, in diff
    return f" ({str(column / row['sqlglot'])[0:5]})"
                    ~~~~~~~^~~~~~~~~~~~~~~~
TypeError: ufunc 'divide' not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule ''safe''

Will have to be something for tomorrow.

[tobymao]

why is this optional?