gh-36422: replace wrongly placed sig_on/off pair with sig_check
fixes the bug noted in https://github.com/mpmath/mpmath/issues/723
The problem comes from the `sig_on/off` pair guarding non-Cython
statements; we replace it by `sig_check()`
```
sage: from mpmath import *
sage: mp.dps=16
sage: zeta(-0.01 + 1000j)
------------------------------------------------------------------------
---
SystemError Traceback (most recent call
last)
Cell In [3], line 1
----> 1 zeta(-RealNumber('0.01') + ComplexNumber(0, '1000'))
File /usr/lib/python3.11/site-packages/mpmath/functions/zeta.py:580, in
zeta(ctx, s, a, derivative, method, **kwargs)
578 if ctx.re(s) > 2*ctx.prec and a == 1 and not derivative:
579 return ctx.one + ctx.power(2, -s)
--> 580 return +ctx._hurwitz(s, a, d, **kwargs)
File /usr/lib/python3.11/site-packages/mpmath/functions/zeta.py:595, in
_hurwitz(ctx, s, a, d, **kwargs)
593 print("zeta: Attempting reflection formula")
594 try:
--> 595 return _hurwitz_reflection(ctx, s, a, d, atype)
596 except NotImplementedError:
597 pass
File /usr/lib/python3.11/site-packages/mpmath/functions/zeta.py:654, in
_hurwitz_reflection(ctx, s, a, d, atype)
652 p += shift*q
653 assert 1 <= p <= q
--> 654 g = ctx.fsum(ctx.cospi(t/2-2*k*b)*ctx._hurwitz(t,(k,q)) \
655 for k in range(1,q+1))
656 g *= 2*ctx.gamma(t)/(2*ctx.pi*q)**t
657 v += g
File /mnt/opt/Sage/sage-dev/src/sage/libs/mpmath/ext_main.pyx:767, in
sage.libs.mpmath.ext_main.Context.fsum()
765 workopts.rounding = ROUND_D
766 unknown = global_context.zero
--> 767 sig_on()
768 try: # Way down, there is a ``finally`` with sig_off()
769 MPF_init(&sre)
SystemError: calling remove_from_pari_stack() inside sig_on()
```
URL: https://github.com/sagemath/sage/pull/36422
Reported by: Dima Pasechnik
Reviewer(s): Dima Pasechnik, Kwankyu Lee
