Rename TokenizerConfig.__post__init__ to __post_init__

[Chessing234]

Bug

In python/dolma/cli/tokenizer.py, TokenizerConfig declares

def __post__init__(self):
    ...
    if self.pad_token_id is None:
        self.pad_token_id = self.eos_token_id
    ...

but the method name has an extra underscore. The @dataclass machinery
only invokes __post_init__ (single underscore between post and
init), so this method is never called.

Root cause

Plain typo in the method name (__post__init__ instead of
__post_init__). Nothing else in the file references it, so the typo
is silent: the code compiles and imports fine, and the tokenizer CLI
just runs without the intended post-init validation.

Why the fix is correct

Renaming to __post_init__ makes @dataclass invoke it exactly as
intended:

• pad_token_id falls back to eos_token_id when the user doesn't
  pass one (line 66). Without the fix, pad_token_id stays None
  and propagates into tokenize_in_parallel(..., pad_token_id=None, ...)
  at line 222 of TokenizerCli.run.
• The "NO EOS TOKEN PROVIDED", "NO BOS TOKEN PROVIDED", and
  "segment_before_tokenization is experimental" warnings fire for
  misconfigured runs, instead of being silently skipped.

This is a one-character change, no logic in the body of the method is
touched, and no call site needs to change because nothing was
explicitly calling the misnamed method.

Change

python/dolma/cli/tokenizer.py: def __post__init__(self): →
def __post_init__(self):.