mpmath.mp.mpf has outdated ``__float__`` dunder

[skirpichev]

One looks like:

sage/src/sage/libs/mpmath/ext_main.pyx

Lines 1784 to 1794 in 31e2166

	def __float__(self):
	"""
	Support float conversion for derived classes ::
	sage: from mpmath import mpf
	sage: from sage.libs.mpmath.ext_main import mpf_base
	sage: class X(mpf_base): _mpf_ = mpf(3.25)._mpf_
	sage: float(X())
	3.25
	"""
	return libmp.to_float(self._mpf_)

i.e. context's rounding mode is ignored and this function fallback to to_float() defaults (i.e. "round down", while context's defaults - "round to nearest"). In the mpmath:

    def __float__(s): return to_float(s._mpf_, rnd=s.context._prec_rounding[1]) 

Reproducer (see mpmath/mpmath#809):

from mpmath import mp, mpf, log
mp.prec=128
print(float(log(mpf('5.3e-7'))))

CPython3.12.4 + mpmath1.3.0 - -14.450388830400243
SageMathCell - -14.450388830400245

Simple fix is - adjust just this dunder. But I suggest switch to use the mpmath's upstread pure-python code. (1) "faster versions" are outdated c.f. the mpmath master, there are other problems, e.g. #36447 (2) using gmpy2.mpfr type as a drop-in replacement for the mpf is incorrect: these types actually are different (mpfr has signed zero, mpf has no limits for exponents range).

[mkoeppe]

I suggest switch to use the mpmath's upstread pure-python code.

@skirpichev This is happening in #38113