Fix: use alpha + tau as the branch threshold in Huber.prox

[raashish1601]

Huber.prox picked the branch with |x| > alpha, but for the prox of tau * H_alpha the switch is at |x| = alpha + tau. Inputs with alpha < |x| <= alpha + tau were soft-thresholded instead of scaled, so the result wasn't the minimizer and the prox wasn't monotone (with alpha = tau = 1, x = 1.2 gave 0.2 while x = 1.0 gave 0.5). See #275.

The fix changes the mask to |x| > alpha + tau. With that, the two branches give the closed form (1 - tau / max(|x|, alpha + tau)) * x that was left commented out in prox, so I kept it as a reference and corrected the comment that said it gave a different result. The Notes section now shows the correct thresholds.

Tests: added test_Huber_prox in pytests/test_norms.py, which compares prox with a brute-force minimizer of tau * H(y) + (y - x)^2 / 2 on a grid of x that covers alpha < |x| <= alpha + tau. It fails on dev and passes here. pytests/test_norms.py passes, and ruff check, ruff format --check and mypy are clean on the changed files.

AI assistance: this PR was prepared with an AI coding agent (Claude Code, Anthropic). It reproduced the issue, wrote the fix and tests, and ran them locally as described above.

Closes #275

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

🟢 Metrics 0 complexity · 0 duplication

Metric	Results
Complexity	0
Duplication	0

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}