Let $u_n=\frac{1}{\log(a_n)}$, we have $\begin{cases}u_{n+1}=f(u_n)\\u_0=x\end{cases}$ where $f(x)=(x^{-1}+\log(1+x))^{-1}$
This function has expansion $f(x)=x+ax^{p}+O(x^{p+1})$ and the sequence converges $u_n\to0$ , thus we adapt a general technique from this answer on MO (recommend reading for explanation, I will just summarize the main result), that is we need to find the expansion of Fatou coordinate $\alpha(x)$ where $\alpha(f(x))=\alpha(x)+1$ , this can be done via expansion of $1/\alpha'(x)$. In general, we have the result:
$$\alpha(x)=\lim_{n\to\infty}\left(\alpha(u_n)-n\right)$$
By symbolic computation:
$$f(x)=(x^{-1}+\log(1+x))^{-1}\implies \alpha(x)\sim\frac{1}{2x^2}+\frac{1}{2x}+\frac{7}{12}\log x+O(1)$$
Let $b_n=1/u_n$, we have:
$$n\sim G(b_n)=\frac{1}{2}b_n^2+\frac{1}{2}b_n-\frac{7}{12}\log b_n+O(1)$$
The idea is that we just substitute $n$ into the limit. The limit in question:
$$\log L=\lim_{n\to\infty}\left(b_n-\sqrt{2n}\right)$$$$=\lim_{n\to\infty}\left(b_n-\sqrt{2\left(\frac{1}{2}b_n^2+\frac{1}{2}b_n-\frac{7}{12}\log b_n+O(1)\right)}\right) $$$$=\lim_{x\to\infty}\left(x-\sqrt{x^2+x-\frac{7}{6}\log x+O(1)}\right)=\frac{-1}{2}$$
The answer is $$L=\lim_{n\to\infty}\frac{a_n}{e^{\sqrt{2n}}}=e^{-\frac{1}{2}}$$
Below is the Mathematica code I wrote to calculate $G(x)$ (or $\alpha(x)$) and also the limit in question:
(* sequence b(n+1)=myFun(b(n)), b(n)->0, myFun(x)=x+O(1) "Laurent series"*)
(* AbelAsymptotic return asymptotic G where n∼G(b(n))+const *)
AbelAsymptotic[H_,n_]:=Module[{F,f,g,const},
F[x_]=Asymptotic[H[y]-y,y->0]/.y->x;
pow = Limit[Log[Abs[F[x]]]/Log[x],x->Infinity];
f[0,x_]=F[x];
Do[f[x_,c_]=f[i-1,x]+cx^(pow+i);
g=Solve[(Asymptotic[f[H[x],c]-f[x,c]D[H[x],x],x->0]/.x->1)==0,c];
const = c/.g[[1]];
f[i,x_]=f[i-1,x]+constx^(pow+i),{i,pow2+n}];
Simplify[Integrate[Normal[Series[1/f[pow*2+n,x],{x,0,n}]],x]/.(x->1/z),z>0] //Expand
];
myFun[x_]:= x+Log[1+1/x]; (* Input function here *)
fun = Function[x,(myFun[1/x])^(-1)];
AbelAsymptotic[fun,3] (expansion accuracy is after parameter "fun")
(End function, below is code to solve limit problem)
funToCal[x_,y_] = x-Sqrt[2*y]; (the limit x=b(n),y=n)
Limit[funToCal[z,AbelAsymptotic[fun,3]],z->Infinity]