gh-84978: expose __float__ dunder method as as_float

[skirpichev]

This is a lightweight alternative for #26827 (an alternate constructor could be more discoverable option). Similar functions could be added for __complex__() and __int__() methods. But, probably, support of float subclasses in __float__() should be removed first (see #109311).

Of course, this is not an equivalent to PyFloat_AsDouble(), which tries to use __index__() dunder if there is no __float__() one. But maybe it could serve at least as a partial solution for the issue?

---

to decide:

• name: as_float vs float (c.f. index)?

---

• Issue: Expose PyFloat_AsDouble at Python level: operator.as_float? #84978

---

📚 Documentation preview 📚: https://cpython-previews--110460.org.readthedocs.build/