Expose --qparams_algorithm CLI arg for quantization parameter selection

[rhn19]

Fixes #226

Problem

quantize_model_() accepts a qparams_algorithm parameter, but there is no CLI argument to pass it through. Users cannot control which algorithm is used for computing quantization parameters during export.

Changes

• Add --qparams_algorithm CLI argument (choices=["affine", "hqq_scale_only"])
• Wire it through all task loaders (causal_lm, masked_lm, asr, multimodal_text_to_text) to quantize_model_()

Usage

optimum-cli export executorch \
   --model google/gemma-3-1b-it \
   --task text-generation --recipe xnnpack \
   --qlinear 8da4w --qparams_algorithm affine \
   --output_dir output/

No behavior change when --qparams_algorithm is not specified.