Fix tokenizer test failure

[zcbenz]

Fix the failed test:

======================================================================
FAIL [0.000s]: test_tokenizers (test_tokenizers.TestTokenizers) (tokenizer='mlx-community/Llama-3.2-1B-Instruct-4bit')
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/runner/actions-runner/_work/mlx-lm/mlx-lm/tests/test_tokenizers.py", line 68, in test_tokenizers
    self.check_tokenizer(tokenizer)
  File "/Users/runner/actions-runner/_work/mlx-lm/mlx-lm/tests/test_tokenizers.py", line 40, in check_tokenizer
    check(tokens)
  File "/Users/runner/actions-runner/_work/mlx-lm/mlx-lm/tests/test_tokenizers.py", line 31, in check
    self.assertEqual(text, expected_text)
AssertionError: '<|begin_of_text|>a,b' != '<|begin_of_text|>a ,b'
- <|begin_of_text|>a,b
+ <|begin_of_text|>a ,b
?                   +

Fix 2 problems:

• Overriding TokenizerWrapper._detokenizer was not working.
• Do not trim space in the test, which expects "a ,b" to reserve its spaces with streaming decoding which is handled as ["a", " ,", "b"] and would have the space trimmed without setting clean_spaces = False.