fix: forward falsy logits processor in mlxlm format_output_type

[serhiizghama]

Closes #1980. MLXLMTypeAdapter.format_output_type used if not output_type to decide whether a logits processor was supplied. That's a truthiness check, so any processor whose __bool__ (or __len__) evaluates to False gets silently dropped and generation runs unconstrained — no error, just wrong output. Only None should mean "no constraint".

Switched to the identity check if output_type is not None, matching transformers.py, llamacpp.py and the other backends.

Added two tests that don't need Apple Silicon (they build the adapter with a mock tokenizer): one for the None case and one for a falsy processor. The falsy-processor test fails on the old code and passes on the fix.