Dev v0.2.3

[mathiassalvas]

• Issues with hist_optim, rescaling and hist_specification in the CLI were fixed.

• Added mentions to the paper and the readthedocs in the documentation.

• Enhanced StimulusMasker — ready for both SHINIER's own masking and experiment stimulus presentation.
  Three ways to build one: direct construction, from_mask() (fit from an existing mask), or from_interactive_mask() (Matplotlib GUI).

Methods:

- generate_mask() — mask as float64 in [0, 1].
- apply_mask(stim, ...) — apply to one image, a batch, or a {name: image} mapping (names preserved on output).
- save_mask(path, ...) — save the raw mask (.npy or preview image); warns if soft edges make it unsafe to reuse.
- save_masked_stim(stim, path, ...) — apply and save to disk.
- interactive_mask(image) — live GUI tuning (sliders, edge-bias toggle, reset).

Notable additions:

- New edge_bias ("center" / "inward") — soft edges never bleed outside the cutoff boundary in "inward" mode.
- mask_type now defaults to "hard", matching SHINIER's convention.
- Batch errors now name the offending stimulus; duplicate output filenames are rejected instead of silently overwritten.
- Renamed for clarity: mask()→generate_mask(), apply()→apply_mask(), save()→save_masked_stim().