LSR1Operator(1) fails the secant equation

Hello all,

I did test the LSR1 operators for myself, and it seems that there is a problem with the LSR1Operator of dimension 1.
In my work, I generate automatically several LSR1Operators, and some of them are of dimension 1.

Here is an example to replicate the issue:
```julia
n = 1
B = LSR1Operator(n)
r = 0.0
for i = 1:100
global r
s = rand(n)
y = rand(n)
push!(B, s, y)
r += norm(B * s - y) #secant equation
end
r #166.39...
```

If you replace `n=1` by `n=2` then `r = 1e-14`.

The problem does not occur with the LBFGSOperators.