llama-quantize: fix tensor-type logic

[theo77186]

Overview

Fix quantization type selection with --tensor-type when the requested type is the same as the base type.

Additional information

For example, when quantizing to Q4_K_M, using --tensor-type ffn_down=Q4_K fails to get applied at all because in the current code, the condition qtype != new_type would evaluate to false, only to get overridden with the llama_tensor_get_type_impl call.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: NO

[theo77186]

superseded by #22572