Fix stain normalizer float dtype detection

[fallintoplace]

Summary

Fix float input detection in image_to_absorbance so normalized floating-point images use the float clipping path.

What changed

• detect float inputs with image.dtype.kind == "f" instead of comparing the dtype object to the string "f"
• add regression coverage for float32 and float64 inputs in test_stain_normalizer.py
• validate the normalized-float case with source_intensity=1.0, including lower-bound and upper-bound clipping

Why

image_to_absorbance receives image.dtype as a dtype object, so the old input_dtype == "f" check was always false.

That sent float images through the integer clipping path. For normalized inputs, especially with source_intensity=1.0, values below 1.0 were clipped incorrectly before the absorbance transform.

Validation

• reviewed the float/int clipping logic in stain_normalizer.py
• added direct unit coverage for image_to_absorbance with float32 and float64 inputs
• ran python3 -m py_compile on the changed source and test files
• could not run the Python unit tests locally because this machine does not have cupy installed

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.