feat(shape): add e<->g ellipticity converters

[cailmdaley]

What

Adds cs_util.shape, a small module parallel to cs_util.size, as the single home for the distortion (e-type) ↔ reduced-shear (g-type) ellipticity conversion used across the UNIONS / ShapePipe stack:

g = e / (1 + sqrt(1 - e^2))      # e_to_g
e = 2 g / (1 + g^2)              # g_to_e

Both functions are component-wise (the WL-native two-component form): the two ellipticity components are scaled by the magnitude ratio, so orientation is preserved. Round objects map to round objects; the small-ellipticity limit reduces to e ≈ 2g.

No version bump. Downstream (shapepipe) will track cs_util @ develop rather than pinning a release, so new helpers like this are available without shipping a new version each time.

Why

ShapePipe's HSM moments store the distortion e while ngmix stores the reduced shear g; downstream (sp_validation) sometimes needs one given the other. Centralising the conversion here means producers and consumers import a single source of truth rather than re-deriving the factors locally — the same move that motivated cs_util.size.

This closes the converter half of Sacha Guerrini's review point on CosmoStat/shapepipe#761 (HSM is e-type, ngmix g-type — name consistently and add a converter).

Tests

cs_util/tests/test_shape.py — known q=1/3 pair (|e|=0.8 ↔ |g|=0.5), axis-ratio closed forms, component-wise round-trips, orientation preservation, zero handling, small-ellipticity limit, array shapes. All pass.

— Claude on behalf of Cail

---

Closes #73 — the e↔g converter single-source-of-truth (a sub-issue of the ngmix v2.0 epic CosmoStat/shapepipe#762). Linked here via the Development sidebar so the issue tracks this PR and closes when it merges.

[sachaguer]

Apparently I do not have write access to the repo. I have a commit related to this PR. @martinkilbinger can you check if I do have write access and I made a mistake when pushing and if not could you add me?

[cailmdaley]

gave you maintainer access @sachaguer

[sachaguer]

Thanks, the PR meets its goals. As for me, it can be merged if all the CD/CI's pass