Add Sieve of Atkin algorithm for efficient prime generation

[AyhamJo7]

This PR adds an implementation of the Sieve of Atkin algorithm to the maths directory.

[priya-sundaram-dev]

Reviewed as part of the maths/ awaiting-reviews triage (#15081).

Verdict: correct algorithm, genuinely new — recommend merge after removing one stray file.

Not a duplicate. We have several Eratosthenes-family sieves (sieve_of_eratosthenes.py, prime_sieve_eratosthenes.py, segmented_sieve.py, odd_sieve.py), but the Sieve of Atkin is a genuinely different algorithm (quadratic-form based, not trial-marking of multiples), so it passes the no-duplicate bar.

Correctness — verified. I diffed the output against an independent Eratosthenes reference for every limit 0–400 plus 1000, 5000, 10007, 50000, 100000: identical prime sets, 0 discrepancies. Edge cases (0, 1, 2, negative → ValueError) all match the doctests.

One change required before merge:

1. Remove pr_description.txt — it's a PR-writeup file that shouldn't be committed to the repo. Only maths/sieve_of_atkin.py should ship. (This is likely also why a reviewer would hesitate; drop it and the PR is a single clean module.)

Optional polish (non-blocking):
2. The docstring's O(n / log log n) refers to the theoretical optimized Atkin; this straightforward list-of-bools implementation is closer to O(n) work with a smaller constant than a segmented/wheel version. A one-line "this is the basic, non-segmented form" note would set reader expectations accurately.

Clean, well-documented, good doctest coverage on the edge cases. Just drop the stray .txt and this is ready.