docs: further elaborate the limitations and the restarting

doc/colvars-refman-main.tex

@@ -6524,7 +6524,7 @@
 \cvsubsubsec{Implementation notes}{sec:colvarbias_opes_notes}
 Compared to the the OPES module in PLUMED \cite{tribello_plumed_2014}, this implementation currently has the following limitations:
 \begin{itemize}
-\item Colvars does not support chaining bias actions in PLUMED, so the Colvars OPES implementation cannot take the biasing energies of other biases, such as harmonic restraints, into account.
+\item Colvars does not support chaining bias actions in PLUMED, so the Colvars OPES implementation cannot take the biasing energies of other biases, such as harmonic restraints, into account for the internal reweighting, which means that there is no way to achieve the same functionality as the \texttt{EXTRA\_BIAS} option in the OPES module in PLUMED. This means that it is not possible to sample a custom target distribution in the Colvars OPES implementation.
 \item Due to the same limitation of Colvars as described above, you cannot collect samples on-the-fly by chaining \texttt{opes{\textunderscore}metad} with a weighted histogram for PMF calculations. Although the PMF calculation via reweighting is implemented into the \refkey{pmf}{opes_metad|pmf} option, the functionality is quite limited. The CVs in \refkey{pmfColvars}{opes_metad|pmf_colvars} must be among the biased ones. In other words, you cannot apply bias on $(\xi_1, \xi_2)$ and compute the PMF along $\xi_3$ by the on-the-fly reweighting using \refkey{pmf}{opes_metad|pmf} (you can only compute the PMF along $\xi_1$, $\xi_2$ or $(\xi_1, \xi_2)$). In addition, you can only specify one set of CVs for estimating the PMF (you cannot apply the OPES bias on $(\xi_1, \xi_2, \xi_3)$ and compute two 2D PMFs along $(\xi_1, \xi_2)$ and $(\xi_2, \xi_3)$ simultaneously using the \refkey{pmf}{opes_metad|pmf} option). However, you can always bypass this limitation by post-hoc reweighting with the trajectories.
 \item Colvars computes the biases in parallel with SMP, so it cannot use multiple threads again for computing the probability and the normalization factor, searching the kernels that are able to compress and updating the neighbor list. As a result, the CPU performance of this implementation may be slower than the PLUMED OPES module.
 \end{itemize}
@@ -6554,6 +6554,8 @@
 \end{tabular}
 \end{table}
 
+The PLUMED options for restarting (\texttt{STATE\_RFILE}, \texttt{STATE\_WFILE} and \texttt{STATE\_WSTRIDE}) are not necessary in Colvars, since Colvars has a unified mechanism to save the information required for restarting in the \texttt{.colvars.state} file.
+
 \cvsubsubsec{General options}{sec:colvarbias_opes_general_options}
 To enable a OPES-based calculation, a \texttt{opes{\textunderscore}metad \{...\}} block must be included in the Colvars configuration file. The \texttt{opes{\textunderscore}metad} block supports the following options:
 \begin{itemize}