Prefer to use composite isogeny in more cases

[user202729]

Note: make sure to review the dependency before this pull request.

The advantage of this pull request is that .isogeny() finishes quickly when the curve order is highly composite, even if you forget to pass algorithm="factored". Example (from doctest):

sage: E = EllipticCurve(GF((60*2^200-1)^2), [1, 0])
sage: P = E.0+E.1
sage: hasattr(P, '_order')
False
sage: E.isogeny(P)
Composite morphism of degree 964162...

The reason why this is never slower than the existing code is the following. Currently when the kernel generator P has unknown order, it uses EllipticCurveIsogeny, which in turn uses __init_from_kernel_point, as you can see it takes time linear in the order of P.

The new code takes time O(√(order of P)) which is negligible compared to that. (Maybe with some extra log factor.)

📝 Checklist

• The title is concise and informative.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes.
• I have updated the documentation and checked the documentation preview.

⌛ Dependencies

#40223

[github-actions]

Documentation preview for this PR (built with commit 2febe78; changes) is ready! 🎉
This preview will update shortly after each push to this PR.