Check for the case of zero variance

When the time series consists of a straight line, `p.Vt <- sctest(efp(Vt ~ ti, h = h, type = type))` returns `NaN`, which then fails down the line. bfast should throw an error in case sd(Vt) == 0, and possibly catch any further errors when `p.Vt` is invalid.

Start at the bfast path containing `p.Vt <- sctest(efp(Vt ~ ti, h = h, type = type))` and reproduce it with a straight-line time series. Trace how the resulting `NaN` reaches later processing; done means the zero-variance case raises a clear error and invalid `p.Vt` is handled. The issue names no file or test, so locating the entry point and adding regression coverage is part of the work.

Written by the indexing model from the issue text.