Bump fortnum to 48f6f9b for the VODE event-continuation fix

[krystophny]

fortnum's vode_integrate_to used to report an event root by writing
state%tn = troot while leaving the Nordsieck array yh expanded about
the internal mesh top, so any integration continued past a root was
relabelled in time by up to one internal step -- silently, since the
trajectory stays smooth and only its clock is wrong. Worst root-time
error over six roots went from 1.65e-01 to 6.8e-10. The default
event tolerance also changed from a fixed 1e-10 absolute in t to
DVODE's relative rule 100*uround*max(|x0|,|x1|); an explicit
event_tol still wins.

I checked whether libneo exercises fortnum's ODE event API. libneo uses
fortnum_special, fortnum_quadrature, fortnum_fft, and
fortnum_status/fortnum_multiroot, but nothing calls
vode_integrate_to or fortnum's event-root detection -- libneo's own
ode_event_t in src/odeint/odeint_allroutines.f90 is a local type,
unrelated to fortnum's ODE solver. This bump is routine here, not
behaviour-changing.

Link: lazy-fortran/fortnum#92

https://claude.ai/code/session_01EHZELf2d2zR4DQWqLPcnkg