smc_samplers.AdaptiveTempering may raise an error before reaching exponent == 1.

[nchopin]

@FrancescaCrucinio found a corner case where smc_samplers.AdaptiveTempering crashes when the tempering exponents (found numerically by root finding) increases too slowly.

It looks like the issue comes from this line:

particles/particles/smc_samplers.py

Line 918 in 9ec5e2f

delta = optimize.brentq(f, 1.0e-12, 1.0 - epn) # secant search

and the left end of the bracketing interval, which is set to $10^{-12}$.

[nchopin]

Quick fix on experimental branch, seems to work, but I'll test it more thoroughly in the coming days.