fft-cnr: FFT-based contrast-to-noise ratio estimation from a single frame

Packaging release that carries the 0.3.2 changes to PyPI. The package code is unchanged from 0.3.2.

Version 0.3.2 was tagged and archived on Zenodo but not published to PyPI: the publish action rejected the metadata version 2.5 that current hatchling writes. This release updates the publish action to v1.14.2 and pins the build backend (hatchling==1.32.4).

Changes since 0.3.1 (see the 0.3.2 notes):

• The generalized-Gaussian path (fit_model="generalized_gaussian") refuses two kinds of degenerate fit and falls back to the peak method: a fitted amplitude more than twice the data's peak-to-peak range, or a width pinned at the profile length. diagnostics["gaussian_fit_rejected"] names the reason. Results computed with 0.3.1 on this path should be re-run. diagnostics["gaussian_fit_params"] is absent on fallback, so read it with .get().
• README: the peak method's baseline margin is quantified against profile length, and the flat-top bias at high CNR is attributed to leakage above the signal/noise boundary (a diagnostic is tracked in #15).

Full changelog: https://github.com/thiebes/fft-cnr/blob/main/CHANGELOG.md